6QQMN331

1,018 + 1,145 + 1,003 = 3,166

 KBS

ARTIFICIAL INTELIGENCE AT
WORK
INDIVIDUAL COURSEWORK
APRIL 2023
Table of Contents

4
Question 2

13
Question 3

22
Question 5

KBS
 Question 2
Exhibit 1: AI Companies in Hollywood - Industry Specializations and Partnerships

Company Size Film Industry Production

Company Year Founded Headquarters
(employees) Specialization Company Partners

Script
Paramount
ScriptBook Writing/Analysis,
2015 Ghent, Belgium 11-50 Pictures,
[N1] Predictive
STX Entertainment,
Analytics (success
Netflix, BET Films
prediction)

Predictive
VaultAI n/a
2015 California, USA 51-200 Analytics
[N2]
(success
prediction)

Disney, Marvel,
Netflix, Apple TV+,
Marz Visual Effects HBO, Warner Bros,
2018 Ontario, Canada 201-500
[N3] (VFX) AMC, NBC,
Universal,
Paramount
Pictures, Hulu

Largo.AI Lausanne, Predictive

2018 11-50 Netflix
[N4] Switzerland Analytics

Predictive
Paramount
Auckland, Analytics
Movio [N5] 2010 51-200 Pictures, Warner
New Zealand (marketing
Bros
promotion)

Cloud-based
Taylor James,
NVIDIA [N6] 1993 California, USA 10,000+ Virtual
Untold Studios
Workstations

4
 Warner Bros
Data and Sony Pictures
Cinelytic [N7] 2017 California, USA 11-50 Predictive Ingenious Media
Analytics STX
Entertainment

Audiovisual
IBM Watson Analysis 20th
2017 New York, USA 10,000+
Media [N8] and Cognitive Century Fox
Computing

Internal Data
Nielsen
and
(Gracenote Studio 1998 California, USA 1,001-5,000 n/a
Predictive
Solutions) [N9]
Analytics

Predictive
Google (Merlin) Analytics 20th
2018 California, USA n/a
[N10] (success Century Fox
prediction)

Visual Effects
Metaphysic
2021 London, UK 11-50 (VFX) Miramax
[N11]

Predictive
Greenlight Analytics
2012 Ontario, Canada 11-50 Kickstarter
Essential [N12] (automated
creating)

Industrial Lights
Visual Effects
& Magic (ILM) 1975 California, USA 1,001-5,000 LucasFilm
(VFX)
[N13]

Geena Davis
Gender Bias Disney
Institute on
2004 California, USA 11-50 Script Analysis Studios
Gender in Media
(content
(GDI) [N14]
management)

5
Introduction

AI companies working with Hollywood studios are changing the way films are made, marketed, and
distributed, and providing new insights and data to support creative decision-making. This report will
explore the data surrounding AI companies working with Hollywood studios, discuss how AI impacts the
film industry, and assess whether AI decreases creativity in Hollywood.

AI's Impact on the Film Industry

Predominantly based in the USA, the companies in Exhibit 1 can be categorized into 4 key domains where
their industry-specialized AI tools have the greatest impact.

Exhibit 2: Distribution of AI companies (Exhibit 1) by impact categories

Movie Promotion and Success Prediction

AI-driven predictive analytics improves post-production workflows, optimizes marketing and distribution
strategies, and enhances audience engagement (Edgerton, 2020). For example, Merlin helped 20th
Century Fox predicts target audiences for ‘Logan’ based on the trailer, offering valuable insights into
marketing campaign effectiveness – a key determinant of box-office success and distribution locations.

Movie Production

AI can improve efficiency and the viewer experience in movie production, ranging from pre-production
tasks such as actor scheduling and location scouting to enhancing lighting during filming (Hooksounds,
2023). For instance, Metaphysic's AI tools creates high-resolution face swaps and de-aging effects in real-
time, eliminating the need for additional VFX work (Takahashi, 2023).

Editing

AI can assist editors to create movie trailers by identifying high-action and high-emotion areas using
algorithms that analyze visuals, audio, and each scene’s composition. For example, IBM’s AI tool identified
suitable clips for ‘Morgan’s’ trailer, saving 10-30 days of manual labor (Johnson 2022).

6
Scriptwriting

Companies like ScriptBook significantly reduce studio costs with its expertise in script analysis, using
machine learning to generate the commercial viability that would otherwise be in the hands of subjective
executives and require long periods of research and deliberation, providing "financial forecasts with an
86% success rate” (Chow, 2020, p. 199; Seth, 2019).

PART B

Has AI increased creativity in the film industry?

Loss of Human Creativity

AI may struggle to replicate human creativity. Studios like Warner Bros and 20th Century Fox using AI tools
in movie production risk product homogenization and loss of originality. Partner companies like Cinelytic
train the predictive analytics AI tools on fixed data sets on more than 95,000 movies and 500,000 actors,
limiting new ideas and hindering creative decision-making (Chandler, 2020; Rose 2020). This may result in
movies with similar styles, content, or themes that lack unpredictable elements of human contribution,
ultimately stifling creativity.

New Creative Possibilities

Conversely, AI can help studios explore new creative ideas and storylines during pre-production.
Predictive analytics tools can encourage studios to change their artistic or stylistic direction. For instance,
Cinelytics’ ability to predict box office success could encourage Warner Bros (Exhibit 1) to explore different
superhero storylines in line with audience sentiment and expectation, rather than produce another
Batman v. Superman (Chandler, 2020; Pallotta, 2022), to increase the probability of success. Therefore, AI
could foster creativity and reduce movie production homogeneity, particularly observable in film series
such as Warner Bros' Justice League.

Risk of Bias

On the other hand, AI may be trained on biased data, leading to a lack of creativity concerning diversity,
equality, and fair representation in movies. ScriptBook’s AI model (Exhibit 1), powered by over 6,500
scripts, could predominantly feature white male protagonists (Caranicas 2018; Chow, 2020), potentially
hindering scripts with different protagonists. Consequently, Paramount Pictures and other production
partners, may limit risk-taking and avoid exploring diverse stories with different perspectives. Thus,
ensuring diverse and inclusive training data is essential to unlocking creative potential in the pre-
production/greenlight stage.

Addressing Bias

Biases may not impact creativity if AI is trained to detect them. Disney’s partnership with GDI (Exhibit 1)
exemplifies how the industry attempts to increase diversity and inclusion with machine-learning tools
(Brzeski, 2019). Therefore, AI-generated output is not always limited by biases, allowing for broader
audience engagement.

7
Facilitating Collaboration

Additionally, AI enhances creativity by enabling artists and editors to work together seamlessly to create
complex visual effects. NVIDIA’s RTX virtual workstations, used by Taylor James (Exhibit 1), accelerate
creative content creation and encourage artists to push the boundaries of visual production while
automating repetitive tasks (Sessions, 2022). These AI tools facilitate efficient cooperation among diverse
teams of artists and content creators, as virtual workstations are accessible from anywhere. Hence, AI
fosters stronger and more efficient collaboration that promotes creativity in the production phase.

Conclusion

Overall, AI is gaining an integral role in the film industry by changing the way production companies
operate. The report has provided deep insights into AI companies that work with Hollywood studios and
their potential to assist filmmakers to make better-informed decisions throughout all stages of
production. However, concerns regarding AI hindering creativity and producing homogenous movies
remain. Assessing whether AI increases creativity in the film industry is a complex and multifaceted
dilemma. Based on the discussion, it is not the AI tool itself that diminishes creativity, but rather the way it
is employed by the production company that may lead to adverse effects. Nevertheless, AI tools can lead
to new creative possibilities and greater collaboration during pre/mid-production. The key to harnessing
the power of AI and ensuring the film industry thrives creatively in this era will require a balance between
technology and human creativity, leveraging AI as a complement rather than a substitute for human input.

8
Notes [N]: Cross-references (CRF)

[N1] Scriptbook industry specialisation and production partners obtained from: [29] CRF [14], [7] & [33]

[N2] Vault AI industry specialisation obtained from: [7] CRF [36] & [33]

[N3] Marz industry specialisation and production partners obtained from: [2] CRF [8]

[N4] Largo industry specialisation and production partners obtained from: [18] CRF [21]

[N5] Movio industry specialization and production partners obtained from: [12] CRF [18], [33], [10] & [20]

[N6] Nvidia industry specialisation and production partners obtained from: [30] CRF [27]

[N7] Cinelytic industry specialisation and production partners obtained from: [37] CRF [6], [33] & [7]

[N8] IBM Watson Media industry specialisation and production partners obtained from: [32] CRF [16]

[N9] Gracenote industry specialisation and production partners obtained from: [11] CRF [1]

[N10] Merlin industry specialisation and production partners obtained from: [23] CRF [37]

[N11] Metaphysic industry specialisation and production partners obtained from: [22] CRF [34]

[N12] Greentlight Essential industry specialisation and production partners obtained from: [33] CRF [19] &
[4]

[N13] ILM industry specialisation and production partners obtained from: [33] CRF [17]

[N14] GDI industry specialisation and production partners obtained from: [33] CRF [3] & [26]

[N1 - N14] All company's founding date, employee and headquarter data obtained from:
https://www.linkedin.com CRF https://www.crunchbase.com

9
References

[1] Advanced Television. (2022, October 11). Gracenote debuts Gracenote Studio Solutions. Retrieved
March 29, 2023, from https://advanced-television.com/2022/10/11/207127/

[2] Ballard, L. (2023, January 19). Marz unveils vanity AI VFX solution for Hollywood. Retrieved March 20,
2023, from https://www.businesswire.com/news/home/20230119005265/en/MARZ-Unveils-Vanity-AI-
VFX-Solution-for-Hollywood

[3] Brzeski, P. (2019, October 05). Geena Davis unveils partnership with Disney to "Spellcheck" scripts for
gender bias. Retrieved March 20, 2023, from https://www.hollywoodreporter.com/movies/movie-
news/geena-davis-unveils-partnership-disney-spellcheck-scripts-gender-bias-1245347/

[4] Calvario, L. (2016, August 06). 'impossible things' trailer: Kickstarter-backed horror film co-written by
Artificial Intelligence. Retrieved March 20, 2023, from https://www.indiewire.com/2016/08/impossible-
things-trailer-kickstarter-horror-film-co-written-artificial-intelligence-1201714007/

[5] Caranicas, P. (2018, July 05). Artificial Intelligence could one day determine which films get made.
Retrieved March 30, 2023, from https://variety.com/2018/artisans/news/artificial-intelligence-hollywood-
1202865540/

[6] Chandler, S. (2020, January 10). Hollywood is using artificial intelligence to pick its next blockbuster.
Retrieved March 20, 2023, from https://www.forbes.com/sites/simonchandler/2020/01/10/hollywood-is-
using-artificial-intelligence-to-pick-its-next-blockbuster/

[7] Chow, P. (2020, July 06). Ghost in the (Hollywood) machine: Emergent applications of artificial
intelligence in the film industry. Retrieved March 19, 2023, from
https://doi.org/10.25969/mediarep/14307

[8] Delfino, D. (2021, December 03). Marz raises $6.5 million in Series A funding to scale operations and
meet demand for visual effects services. Retrieved March 30, 2023, from
https://monstersaliensrobotszombies.com/blog/marz-raises-65-million-in-series-a-funding

[9] Edgerton, G. (2020). Silicon Valley + hollywood = convergence 2.0. Retrieved March 19, 2023, from
https://digitalcommons.butler.edu/ccom_papers/230/

[10] Eliaçık, E. (2022, November 25). Movio ai narrates your stories with lifelike AI avatars capable of
convincing Lip Sync, gestures and expressions. Retrieved March 29, 2023, from
https://dataconomy.com/2022/11/movio-ai-spokesperson-alternatives-free/?utm_content=cmp-true

[11] Gracenote. (2022, October 11). Nielsen's Gracenote drives next evolution of Hollywood Project
Management with debut of New Studio Solutions Suite. Retrieved March 29, 2023, from
https://www.prnewswire.com/news-releases/nielsens-gracenote-drives-next-evolution-of-hollywood-
project-management-with-debut-of-new-studio-solutions-suite-301645654.html

[12] Heygen. (2023). All in one AI video generator - heygen.com. Retrieved March 30, 2023, from
https://www.heygen.com/article/ai-video-generator?from=moviola

10
[13] Hooksounds. (2023). Artificial intelligence use in the film industry. Retrieved March 30, 2023, from
https://www.hooksounds.com/blog/artificial-intelligence-use-film-industry/

[14] IMDb Pro. (n.d.). Retrieved March 27, 2023, from https://pro.imdb.com/title/tt7737786/?
rf=cons_tt_atf&ref_=cons_tt_atf

[15] Johnson, E. (2022, May 23). Ai's impact on Hollywood. Retrieved March 30, 2023, from
https://aijourn.com/ais-impact-on-hollywood%EF%BF%BC/

[16] Kaufman, D. (2016, September 08). IBM uses AI to create trailer for 20th Century Fox's 'Morgan'.
Retrieved March 29, 2023, from https://www.etcentric.org/ibm-uses-ai-to-create-trailer-for-20th-century-
foxs-morgan/

[17] Kaufman, D. (2019, April 01). Global VFX: State of the industry 2019. Retrieved March 20, 2023, from
https://www.vfxvoice.com/global-vfx-state-of-the-industry-2019/

[18] Largo.ai. (2023). Do you want to join Largo? Retrieved March 20, 2023, from
https://home.largo.ai/about-us/

[19] Ledingham, S. (2016, July 26). Production Company Greenlight Essentials launches first film to use Ar.
Retrieved March 29, 2023, from https://www.prweb.com/releases/2016/07/prweb13571154.htm

[20] Liao, R. (2022, November 08). Movio wants to make your marketing videos with Generative AI.
Retrieved March 29, 2023, from https://techcrunch.com/2022/11/08/movio-marketing-videos-
generative-ai/

[21] Macaulay, T. (2020, November 20). This AI is unimpressed by Gillian Anderson's casting in the Crown.
Retrieved March 30, 2023, from https://thenextweb.com/news/this-ai-is-unimpressed-by-gillian-
andersons-casting-in-the-crown

[22] Metaphysic. (2023, January 31). Metaphysic announces strategic partnership with Creative Artists
Agency; named Generative AI provider for Miramax's "here". Retrieved March 29, 2023, from
https://www.prnewswire.com/news-releases/metaphysic-announces-strategic-partnership-with-creative-
artists-agency-named-generative-ai-provider-for-miramaxs-here-301734767.html

[23] Oakley, S., & Campo-Rembado, M. (2018). How 20th century Fox uses ML to predict a movie audience
| google cloud blog. Retrieved March 20, 2023, from https://cloud.google.com/blog/products/ai-machine-
learning/how-20th-century-fox-uses-ml-to-predict-a-movie-audience/

[24] Pallotta, F. (2022, July 29). Why movie theaters show so few films now | CNN business. Retrieved
March 30, 2023, from https://edition.cnn.com/2022/07/29/media/movie-box-office-year/index.html

[25] Posts, L. (2020). Largo.ai analysed the cast of netflix original: The Crown. Retrieved March 20, 2023,
from http://largofilms.ch/largo-ai-analysed-the-cast-of-netflix-original-the-crown/

11
[26] Pulver, A. (2019, October 09). Geena Davis announces 'spellcheck for bias' tool to redress gender
imbalance in movies. Retrieved March 20, 2023, from
https://www.theguardian.com/film/2019/oct/09/geena-davis-institute-gender-media-disney-machine-
learning-tool

[27] Reid, S. (2022). Untold studios: Pioneering the studio in the cloud: Nvidia on-demand. Retrieved
March 20, 2023, from https://www.nvidia.com/en-us/on-demand/session/siggraph2022-sigg22-s-05/?
playlistId=playList-a4a34631-85f3-45bd-9e8f-d7bd02cd0ec1

[28] Rose, S. (2020, January 16). 'it's a war between technology and a donkey' – how ai is shaking up
Hollywood. Retrieved March 20, 2023, from https://www.theguardian.com/film/2020/jan/16/its-a-war-
between-technology-and-a-donkey-how-ai-is-shaking-up-hollywood

[29] ScriptBook. (n.d.). Scriptbook. Retrieved March 20, 2023, from

https://www.scriptbook.io/#!/scriptbook

[30] Sessions, N. (2022, November 15). Production studio Taylor James elevates remote workflows with
Nvidia Technology. Retrieved March 20, 2023, from https://blogs.nvidia.com/blog/2022/01/10/taylor-
james-remote-work-vgpu/

[31] Seth, R. (2019, June 13). What new technology means for the future of film. Retrieved March 21,
2023, from https://www.vogue.co.uk/article/what-vr-ai-and-interactivity-mean-for-the-future-of-film

[32] Smith, J. R. (2016, November 28). IBM research takes Watson to Hollywood with the first "Cognitive
movie trailer". Retrieved March 20, 2023, from https://www.ibm.com/blogs/think/2016/08/cognitive-
movie-trailer/

[33] Song, M. (2016). A study on the predictive analytics powered by the artificial intelligence in the movie
industry. Retrieved March 20, 2023, from https://koreascience.kr/article/JAKO202106355568899.page

[34] Takahashi, D. (2023, January 31). Metaphysic partners with CAA to develop Generative AI for Creative
Artists. Retrieved March 29, 2023, from https://venturebeat.com/games/metaphysic-partners-with-caa-
to-develop-generative-ai-for-creative-artists/

[35] Vanity AI - the world's first end-to-end AI solution for Hollywood VFX. (n.d.). Retrieved March 20,
2023, from https://monstersaliensrobotszombies.com/vanityai

[36] VaultAI. (2023). Unlock the future. Today. Retrieved March 29, 2023, from https://vault-ai.com/

[37] Vincent, J. (2018, November 02). 20th Century Fox is using AI to analyze movie trailers and find out
what films audiences will like. Retrieved March 20, 2023, from
https://www.theverge.com/2018/11/2/18055514/fox-google-ai-analyze-movie-trailer-predict-success-
logan

12
Question 3
Part A

Exhibit 1: Typology [N1]

Investment Governance Social

Pooling funds to invest in Managing decentralized

Networking and coordinating
crypto, startups, or other organization and on-chain
with a community
assets management of a community

DeFi, Protocol, Media, Grant DeFi, Protocol, Grant Media, Entertainment

Yearn, Olympus, UniSwap,

Olympus, FlamingoDAO, AAVE,
CompoundDAO, LexDAO, BAYCDAO, Fluffworld, FWB,
MetaCartel, AAVEGrants,
CookFinance, GitcoinDAO, Bankless, Forefront
MakerDAO
CookFinance

Art Gaming Philanthropy

Supporting artists and art

Developing and managing Raising and distributing funds
projects, managing art
decentralized games and for charitable causes and
collections, and facilitating the
gaming platforms public goods
buying and selling of art

DeFi, Collector Protocol, Entertainment DeFi, Grant

Sandbox,
NTFfiDAO, ArtsDAO, Giveth, MolochDAO,
DecentralizedGamesDAO,
PleasrDAO BigGreenDAO
MeritCircle

The typology exhibits DAOs by six domains of activity and has subcategorized DAOs in their respective
domain by ‘category’ (Cointelegraph, 2022). Exhibit 2 provides a framework for the typology to illustrate
the non-mutual exclusivity of domains of activity and categories to better understand the potential
applications and disadvantages of DAOs discussed in Part B. The classifications can be subjective due to
the non-mutual exclusive nature of DAOs.

13
Exhibit 2: Sub-Typology [N2]

Category
DeFi Protocol Media Entertainment Collector Grant
Activity

FLAMINGODAO

Investment MAKERDAO

OLYMPUSDAO METACARTEL

Governance COOKFINANCE

GITCOIN DAO

BAYCDAO BANKLESS
Social
FOREFRONT

FWBDAO

Gaming

SANDBOX DECENTRALIZEDGAMES
DAO

ARTSDAO
Art
NTFIfiDAO

PLEASRDAO

MOLOCHDAO
Philanthropy
GIVETHDAO

BIG-GREENDAO

14
 Growth Trends of DAOs

Exhibit 3: [N3]

Exhibit 4: [N3]

Exhibits 3 and 4 indicate how DAO's market capitalization and cumulative investment returns on tokens
have been growing since 2016. The two notable bull markets (Jan-18 & Jan-21) can be attributed to a rise
in investment that was fueled by the ‘fear of missing out’, a social phenomenon resulting from high media
attention (Thurman, 2021). Nevertheless, the cumulative returns, between 100-160% from January 2016
to 2022, indicates the growing financial appeal of DAO tokens as an untraditional asset class to investors.

15
Exhibit 5: [N4]

The above data concentrates on DAOs in the ARBITRUM, MAINNET, POLYGON, and XDAI networks.

The charts present how there were around 2,344 active DAOs and 26,000 active members, implying that
the average DAO is primarily impacted by 11 unique people (Bellavitis, Fisch, & Momtaz, 2022). Although
still a relatively small number, this has been increasing since 2019 highlighting how decentralized
communities in all domains are growing. Furthermore, data indicates that as of February 2022, DAOs
received approximately 18 proposals each and had 8,500 registered voters in the combined networks. This
rapid increase in governance activity suggests a high level of confidence in the DAO structure.

Part B

Decentralized governance ensures accountability and transparency

DAO's employ a flat hierarchy, distributing decision-making power among participants based on their
contributions or stakes (Gogel, Werbach, Slavin, & Kremer, 2022). This decentralization promotes social
community building through philanthropic initiatives, such as BigGreenDAO (Typology), as voting
streamlines bureaucratic grant processes (Wilser, 2022). Decentralized governance also facilitates
innovation, allowing game developers and artists to experiment freely and provide original contributions
with no fear of repercussion (Bellavitis, Fisch, & Momtaz, 2022; Chen & Bellavitis, 2020).
Additionally, Exhibit 6 reveals that governance in most DAOs has low ownership thresholds, making DAOs
suitable for pooled investments, as decision makings is increasingly conducted by a greater proportion of
members (Exhibit 5). By giving investors a say in the organizational direction, the transparent and
accountable system minimizes corruption and bureaucracy, fostering trust among members.

16
Exhibit 6: Ownership threshold for proposals [N5]

While DAOs are purportedly hierarchically flat, Exhibit 6 shows that 15% of the aggregated DAOs maintain
ownership thresholds for members to submit proposals. For instance, CookFinance (Typology) requires
holding almost $1,500 in COOK tokens to propose (Coinmarketcap, 2023). This may result in a small group
controlling the organization, potentially reintroducing hierarchy into DAOs.

Susceptibility to malicious actors and security risks

Additionally, DAOs can be vulnerable to security risks such as hacking or exploitation. For example, $60
million in funds invested in the project AnubisDAO through token sales vanished, causing significant
investor losses (Team, 2022). This can render pooling money for investment or philanthropic donations in
DAOs unsafe and expose participants to scams. Security risks may also endanger gamers through smart
contract exploitation in in-game transactions. Malicious actors have also compromised the integrity of
DAOs in the Social and Art domains due to influencers and artists promoting DAO projects that facilitate
‘rug pulls’ (theft) (Patterson, 2022). Consequently, security risks can erode trust among members and
voting mechanisms, leading to inefficiencies in DAO governance processes. This could undermine the
trend of increasing confidence in the DAO structure observed in Exhibit 5.

Reduced barriers to entry and global collaboration

Nevertheless, another advantage of DAOs is their potential to lower barriers to entry and promote global
collaboration by democratizing access to opportunities. Venture-style investments through crowdfunding
provide access once reserved for wealthy individuals or large institutions (Wright, 2021). DAOs can also
facilitate global cooperation on philanthropic projects as individuals from different backgrounds to
address social issues and initiatives. In the Art and Gaming domains, DAOs enable fractional ownership of
artworks and foster collaborative game development among gamers, developers, and designers
worldwide, facilitating collective decision-making.

Lack of legal recognition and regulatory uncertainty

The legal status of DAOs remains uncertain in many jurisdictions, posing potential compliance and
enforcement challenges that hinder widespread adoption. In the Gaming and Art domains, participants
may face intellectual property rights issues, limiting partnership opportunities (Edwin Coe LLP, 2022).
Without clear legal recognition, DAOs might struggle to attract investment or make philanthropic

17
donations as non-profit organizations due to tax implications (A16Z, 2021). Thus, regulatory uncertainty
can restrict the growth and scalability of DAOs as alternative governance models. The absence of a legal
framework could partly explain the stagnation observed in DAOs market capitalization, as investment is
deterred (Exhibit 3).

Exhibit 7: Summary

Conclusion

DAOs represent a transformative shift across various domains, but their full potential remains unrealized
due to their early developmental stage. By channeling resources efficiently towards innovative
entrepreneurs and worthy causes, DAOs foster innovation and creativity in Art, Gaming, and Social
domains through global collaboration. However, challenges such as the lack of legal recognition and
governance issues need addressing for their success. Ultimately, the long-term impact of DAOs will
depend on how effectively they resolve these challenges, determining whether they can yield significant
economic and social gains.

18
Notes [N]: Cross-references (CRF)

[N1] Typology constructed from: [10] & [8] CRF [2] & [11]

[N2] Sub-typology categorisation obtained from: [2] CRF [8]

[N3] Market Capitalization and Cumulative returns obtained from: [4] CRF [6] & [7]

[N4] Cumulative: proposals, DAOs with registered activity, voters, members obtained from: [4] CRF [3]

[N5] Ownership threshold obtained from [4] CRF [3]

19
References

[1] A16Z. (2021). A legal framework for decentralized autonomous organizations - A16Z crypto. Retrieved
March 25, 2023, from https://api.a16zcrypto.com/wp-content/uploads/2022/06/dao-legal-framework-
part-1.pdf

[2] Accubits. (2022, April 18). What are the different types of daos? Retrieved March 25, 2023, from
https://accubits.com/insights/what-are-the-different-types-of-daos/

[3] Arroyo, J., Davó, D., Faqir-Rhazoui, Y., & Martínez Vicente, E. (2023, February 23). Dao Analyzer.
Retrieved April 9, 2023, from https://doi.org/10.5281/zenodo.7669689

[4] Bellavitis, C., Fisch, C., & Momtaz, P. P. (2022, August 31). The rise of Decentralized Autonomous
Organizations (daos): A first empirical glimpse. Retrieved March 23, 2023, from
https://www.tandfonline.com/doi/full/10.1080/13691066.2022.2116797

[5] Chen, Y., & Bellavitis, C. (2020, June). Blockchain disruption and decentralized finance: The rise of
decentralized business models. Retrieved March 27, 2023, from
https://www.sciencedirect.com/science/article/pii/S2352673419300824

[6] Coingecko. (n.d.). Cryptocurrency prices, charts, and crypto market cap. Retrieved March 29, 2023,
from https://www.coingecko.com/

[7] Coinmarketcap. (2023). Cook finance price Today, Cook to usd live, marketcap and Chart. Retrieved
March 27, 2023, from https://coinmarketcap.com/currencies/cook-protocol/

[8] Cointelegraph. (2022, September 23). Types of daos and how to create a decentralized autonomous
organization. Retrieved March 25, 2023, from https://cointelegraph.com/daos-for-beginners/types-of-
daos

[9] Edwin Coe LLP. (2022, September). Ownership of IP rights by Daos – the future is nigh? Retrieved
March 28, 2023, from https://www.edwincoe.com/ownership-of-ip-rights-by-daos-the-future-is-nigh/

[10] Gogel, D., Werbach, K., Slavin, A., & Kremer, B. (2022, June). Decentralized Autonomous
Organizations: Beyond the hype. Retrieved March 25, 2023, from
https://www.weforum.org/whitepapers/decentralized-autonomous-organizations-beyond-the-hype

[11] Gondek, C. (n.d.). 10 examples of Decentralized Autonomous Organizations (daos). Retrieved March
25, 2023, from https://originstamp.com/blog/10-examples-of-decentralized-autonomous-organizations/

[12] Jentzsch, C. (2017). Decentralized Autonomous Organization to Automate Governance. Retrieved

March 25, 2023, from http://cryptochainuni.com/wp-content/uploads/Decentralized-Autonomous-
Organization-To-Automate-Governance.pdf

20
[13] Patterson, D. (2022, March). New Tech, old scams: Don't fall for these crypto and NFT ripoffs.
Retrieved March 27, 2023, from https://www.cbsnews.com/news/cryptocurrency-nft-scams/

[14] Rohr, J., & Wright, A. (2017). Blockchain-based token sales, initial coin offerings, and the
democratization of Public Capital Markets. SSRN Electronic Journal. doi:10.2139/ssrn.3048104

[15] Team, I. (2022, August 02). How regulators are responding to rising rates of defi fraud. Retrieved
March 25, 2023, from https://integrityriskintl.com/regulatory-response-to-defi-fraud/

[16] Thurman, A. (2021, September 28). Daos may be the future of work, but don't bet on them being
the next big asset class. Retrieved March 28, 2023, from
https://www.coindesk.com/business/2021/09/26/daos-may-be-the-future-of-work-but-dont-bet-on-
them-being-the-next-big-asset-class/

[17] Wilser, J. (2022, June 02). Kimbal Musk and his big green dao. Retrieved March 27, 2023, from
https://www.coindesk.com/business/2022/06/02/kimbal-musk-and-his-big-green-dao/

[18] Wright, A. (2021, June 30). The rise of decentralized autonomous organizations: Opportunities and
challenges. Retrieved March 23, 2023, from https://stanford-jblp.pubpub.org/pub/rise-of-
daos/release/1

21
Question 5
Introduction

Synthetic data, which mimics the properties of real data without being real data itself, is increasingly being
generated for training and building artificial intelligence and machine learning (AI/ML) models across an
array of domains (Emam, Mosquera, & Hoptroff, 2020). This discussion will focus on the application of
synthetic data in medical imaging and simulation-based autonomous vehicle training/testing, seeking to
explain how it may be better than human-generated data. The following discussion combines parts A and
B under each application.

Medical Imaging

One practical example is blood vessel segmentation, separating blood vessels from the background and
other structures in an image. This provides valuable information for diagnosing and monitoring diseases
that affect the blood vessels, such as diabetic retinopathy (Ananthi, 2021).

A study on using synthetic images for training retinal blood vessel segmentation ML models found that,
while not identical, the synthetic images closely resembled real images and had a more consistent level of
similarity (mean image synthesis comparison score close to 1). However, they exhibited less
diversity/variability (standard deviation closer to 0) compared to real images (Exhibit 2). The similarity
between synthetic (left) and real images (right) can be seen in Exhibit 1 (Costa et al., 2018).

Exhibit 1: Synthetic and Real Data [N1]

Exhibit 2: Statistical Comparison [N2]

Improves machine learning model accuracy

By training the machine-learning model on a larger set of synthetic data, the probability of a successful
diagnosis increases (Dilmegani, 2021). As aforementioned, blood vessel segmentation can indicate
diabetic retinopathy, but the machine learning tools are only as good as the data they are trained on.

22
Augmenting the data set with synthetic data can overcome issues such as data insufficiency, under-
represented medical conditions, or privacy concerns that often limit the availability of real-world data in
medical imaging (Fawaz, Forestier, Weber, Idoumghar, & Muller, 2018; Koperniak, 2017; Chen, Lu, Chen,
Williamson, & Mahmood, 2021). Synthetic images, which closely resemble real images can help create
more robust representations of underlying patterns in the data as they deviate less than a set of real
images (exhibit 2), enhancing the segmentation model's ability to recognize abnormalities.

Additionally, synthetic data can be employed to create ground truth labels or annotations, enabling more
effective evaluation and validation of the models (Gao et al., 2023). By addressing these challenges,
synthetic data can significantly improve the generalization capabilities of machine learning models in the
medical imaging domain.

Limitations in capturing rare diseases

However, an alternative interpretation to exhibit 2 reveals limitations of synthetic data in the medical
imaging domain. The relatively, small standard deviation observed for synthetic images indicates less
variability in the data set compared to real-world data. This could lead to the omission of outliers, which
are critical for capturing rare diseases such as fibromuscular dysplasia - a blood vessel disease found
predominantly in middle-aged women. In such cases, outliers in the data may be more important than
other data points (Dilmegani, 2018; John Hopkins Medicine, 2021).

Simulation Based Training for Autonomous Vehicles (AV)

Exhibit 3: AI Companies for AV

Company AI/ML Training or Testing Use Case

“Carcraft” simulates diverse

scenarios for training and
Waymo [N3] Both
validating AV perception models
with synthetic data

Simulates driving conditions in

Cruise [N4] Training
the “Matrix” platform to train AV
systems with synthetic data

Combines synthetic with real data

input to improve AI model
Zoox [N5] Both
navigation of "edge cases" –
rare/complex scenarios

23
 ADAS simulation creates synthetic data
Cognata [N6] Both to support development of partner
company AV systems

DRIVE Sim generates synthetic data for

NVIDIA [N7] Training training AI models on decision-making
and control systems

FaceAPI service generates synthetic data

SynthesisAI [N8] Both of facial expressions for driver safety
monitoring systems

"Waabi World" is a simulator that

Waabi [N9] Both generates synthetic data and trains and
teaches the "brain" to drive safely

Larger amounts of Data at lower costs

Synthetic data can be generated at lower costs and in larger volumes compared to collecting real-world
data. Utilizing driving simulations, companies can generate required datasets on-demand without being
constrained by physical limitations like car technical issues, fuel requirements, or driver fatigue (Toews,
2020). For instance, Waymo (Exhibit 1) generated 2.5 billion miles of simulated driving compared to only 3
million real-world miles for AI system training (Madrigal, 2018). The absence of physical obstacles in
synthetic data generation makes it a more efficient means of gathering than real data. Real-world data
may also contain personal information, rendering it unusable or difficult and costly to collect. Enormous
datasets are required to train AI models, and synthetic data helps overcome the limitations of real data
(Toews, 2022).

Synthetic Data for Safer and Comprehensive Training of AI Models

Another advantage of synthetic data use in simulation-based training for autonomous vehicles is its ability
to cover rare or dangerous situations that are challenging or risky to obtain with real-world data.
Achieving autonomous driving safety requires vast amounts of training hours in various scenarios. For
instance, Waymo, Cruise, and Waabi (Exhibit 1) train their perception models using synthetic data for
driving speeds and control levels unattainable with human-generated data (Bellan, 2022). To train an AI

24
to avoid car crashes, training data on crashes are needed which is not feasible to gather through real
crashes due to its cost and injury hazard; simulated car crashes are performed instead (Naber, 2023).
Consequently, generative models trained with synthetic data are better prepared for real-world
unpredictability compared to those trained with human-generated data, as they are tested under diverse
weather conditions and various parameters, such as different vehicles, times of day, and human behavior
(Team, 2022).

Conclusion

Overall, synthetic data is instrumental in improving machine learning models in both medical imaging and
the training/testing of autonomous vehicles. By addressing data insufficiency, under-representation, and
privacy concerns, synthetic data allows for more robust training of AI models. In the case of medical
imaging, this may contribute to increased longevity in humans if diseases can be identified faster.
Additionally, synthetic data enables safe and efficient explorations of rare and dangerous scenarios,
particularly in the context of autonomous vehicles. In comparison to human-generated data, synthetic
data effectively overcomes many limitations and facilitates greater volumes of data sets to be used in
conjunction with AI/ML models. Therefore, in these domains, synthetic data will likely continue to make
up a greater proportion of data used in AI/ML models in the future.

25
Notes [N]: Cross-references (CRF)

[N1] Synthetic and real pictures obtained from: [5] CRF [17]

[N2] Statistics on synthetic and real data obtained from: [5] CRF [17]

[N3] Waymo profile obtained from: [24] CRF [2] & [21]

[N4] Cruise profile obtained from: [2] CRF [21]

[N5] Zoox profile obtained from: [25] CRF [21] & [16]

[N6] Cognata profile obtained from: [4] CRF [21]

[N7] NVIDIA profile obtained from: [6] CRF [22]

[N8] Synthesis AI profile obtained from: [18] CRF [22]

[N9] Waabi profile obtained from: [23] CRF [2]

26
References

[1] Ananthi. (2021, November 30). Identification of diabetic retinopathy using deep learning algorithm and
blood vessel extraction. Retrieved March 30, 2023, from
https://www.openaccessjournals.com/articles/identification-of-diabetic-retinopathy-using-deep-learning-
algorithm-and-blood-vessel-extraction-
15126.html#:~:text=Diabetic%20retinopathy%20is%20found%20using,method%20for%20blood%20vessel
%20extraction.

[2] Bellan, R. (2022, November 16). Parallel domain says autonomous driving won't scale without synthetic
data. Retrieved April 2, 2023, from https://techcrunch.com/2022/11/16/parallel-domain-says-
autonomous-driving-wont-scale-without-synthetic-data/

[3] Chen, R., Lu, M., Chen, T., Williamson, D., & Mahmood, F. (2021, June 15). Synthetic data in Machine
Learning for medicine and Healthcare. Retrieved March 29, 2023, from
https://www.nature.com/articles/s41551-021-00751-8#ref-CR1

[4] Cognata. (2022, September 13). Sensor Simulation - Autonomous Driving & ADAS Simulation Platform
%. Retrieved March 27, 2023, from https://www.cognata.com/sensor-simulation/

[5] Costa, P., Galdran, A., Meyer, M. I., Niemeijer, M., Abramoff, M., Mendonca, A. M., & Campilho, A.
(2018). End-to-end adversarial retinal image synthesis. IEEE Transactions on Medical Imaging, 37(3), 781-
791. doi:10.1109/tmi.2017.2759102

[6] Cragun, M. (2022, September 13). Nvidia Omniverse Replicator accelerates AV Development. Retrieved
March 27, 2023, from https://blogs.nvidia.com/blog/2021/11/09/drive-sim-replicator-synthetic-data-
generation/

[7] Dilmegani, C. (2018). What is synthetic data? use cases & benefits in 2023. Retrieved March 27, 2023,
from https://research.aimultiple.com/synthetic-data/

[8] Dilmegani, C. (2021). Synthetic data for Healthcare: Benefits & Case Studies in 2023. Retrieved March
29, 2023, from https://research.aimultiple.com/synthetic-data-healthcare/

[9] Emam, E. K., Mosquera, L., & Hoptroff, R. (2020). Practical Synthetic Data Generation: Balancing
Privacy and the broad availability of data. Beijing: O'Reilly.

[10] Fawaz, H., Forestier, G., Weber, J., Idoumghar, L., & Muller, P. (2018, August 07). Data augmentation
using synthetic data for time series classification with deep residual networks. Retrieved March 28, 2023,
from https://arxiv.org/abs/1808.02455

[11] Gao, C., Killeen, B., Hu, Y., Grupp, R., Taylor, R., Armand, M., & Unberath, M. (2023, March 20).
Synthetic data accelerates the development of generalizable learning-based algorithms for X-ray image
analysis. Retrieved March 28, 2023, from https://www.nature.com/articles/s42256-023-00629-1

[12] John Hopkins Medicine. (2021, August 08). Fibromuscular Dysplasia (FMD). Retrieved April 7, 2023,
from https://www.hopkinsmedicine.org/health/conditions-and-diseases/fibromuscular-dysplasia-fmd

27
[13] Koperniak, S. (2017). Artificial data give the same results as real data - without compromising privacy.
Retrieved March 28, 2023, from https://news.mit.edu/2017/artificial-data-give-same-results-as-real-data-
0303

[14] Madrigal, A. (2018, December 04). Waymo built a secret world for self-driving cars. Retrieved March
30, 2023, from https://www.theatlantic.com/technology/archive/2017/08/inside-waymos-secret-testing-
and-simulation-facilities/537648/

[15] Naber, M. (2023, January 27). The advantages of synthetic data over Real Data. Retrieved March 23,
2023, from https://neptune.ai/blog/the-advantages-of-synthetic-data-over-real-data

[16] Sharma, S. (2022, July 28). How edge data is training AI for accurate, real-time response. Retrieved
March 27, 2023, from https://venturebeat.com/ai/how-edge-data-is-training-ai-for-accurate-real-time-
response/

[17] Staal, J., Abramoff, M., Niemeijer, M., Viergever, M., & Van Ginneken, B. (2004). Ridge-based vessel
segmentation in color images of the retina. IEEE Transactions on Medical Imaging, 23(4), 501-509.
doi:10.1109/tmi.2004.825627

[18] Synthesis AI. (2023, March 30). Driver monitoring - synthesis AI. Retrieved March 27, 2023, from
https://synthesis.ai/applications/driver-monitoring/

[19] Team, G. (2022, May 13). Synthetic-data-can-train-safer-self-driving-cars. Retrieved March 23, 2023,
from https://gemmo.ai/mits-research-could-help-engineer-safer-self-driving-cars

[20] Toews, R. (2020, September 29). The peace dividends of the Autonomous Vehicle Wars. Retrieved
March 31, 2023, from https://www.forbes.com/sites/robtoews/2020/08/30/the-peace-dividends-of-the-
autonomous-vehicle-wars/?sh=1999650e2b2c

[21] Toews, R. (2022, October 12). Synthetic data is about to transform artificial intelligence. Retrieved
March 31, 2023, from https://www.forbes.com/sites/robtoews/2022/06/12/synthetic-data-is-about-to-
transform-artificial-intelligence/?sh=37327c575238

[22] Tyrrell, J. (2022, October 18). Synthetic data for AI fills gaps in autonomous vehicle edge cases.
Retrieved March 27, 2023, from https://techhq.com/2022/10/synthetic-data-for-ai-fills-gaps-in-
autonomous-vehicle-edge-cases/

[23] Waabi Blog. (2022, May 09). Welcome to Waabi World. Retrieved March 27, 2023, from
https://waabi.ai/waabi-world/

[24] Waymo. (n.d.). Waymo driver. Retrieved April 7, 2023, from https://waymo.com/waymo-driver/

[25] Zoox. (n.d.). The 'full-stack' behind autonomous driving. Retrieved March 27, 2023, from
https://zoox.com/autonomy/

28