Internet-connected consumer goods are poised for renewed

growth, thanks to the advent of generative AI alongside

hardware and networking technology innovations.

Housetraining robot
Produced in association with
dogs: How generative
AI might change
consumer IoT
2  MIT Technology Review Insights

Key takeaways

1
The convergence of advances in generative
AI, IT hardware, and networking technology
means that new and powerful Internet of
Things (IoT) consumer devices will
increasingly emerge.

A 2
s technology goes, the Internet of Things
This technological capacity will dramatically
(IoT) is old: Internet-connected devices
increase the utility and functionality of
outnumbered people on earth around 2008
existing smart devices and traditional
or 2009, according to a contemporary
appliances. It will also allow innovative new
Cisco report. Since then, IoT has grown
products.
rapidly. Researchers say that by the early 2020s,
estimates of the number of devices ranged anywhere

3
Rapid integration of generative AI and
from the low tens of billions to over 50 billion.
consumer IoT is unlikely due to cost,
problematic attributes of generative AI, a
Currently, though, IoT is seeing unusually intense new
lack of common implementation standards,
interest for a long-established technology, even one
and concerns over safety and data privacy
still experiencing market growth. A sure sign of this
in a context of highly limited regulation.
buzz is the appearance of acronyms, such as AIoT
and GenAIoT, or “artificial intelligence of things” and
“generative artificial intelligence of things.” What is
going on? Why now? IoT, however, is much more. It is a huge technological
ecosystem that encompasses and empowers these
Examining potential changes to consumer IoT could devices. This ecosystem is multi-layered, although
provide some answers. Specifically, the vast range of no single agreed taxonomy exists. Most analyses
areas where the technology finds home and personal will include among the strata the physical devices
uses, from smart home controls through smart themselves (sensors, actuators, and other machines
watches and other wearables to VR gaming—to name with which these immediately interact); the data
just a handful. The underlying technological changes generated by these devices; the networking and
sparking interest in this specific area mirror those in communication technology used to gather and send
IoT as a whole. the generated data to, and to receive information from,
other devices or central data stores; and the software
Rapid advances converging at the applications that draw on such information and other
edge possible inputs, often to suggest or make decisions.
IoT is much more than a huge collection of “things,” The inherent value from IoT is not the data itself, but
such as automated sensing devices and attached the capacity to use it in order to understand what is
actuators to take limited actions. These devices, of happening in and around the devices and, in turn, to
course, play a key role. A recent IDC report estimated use these insights, where necessary, to recommend
that all edge devices—many of them IoT ones—account that humans take action or to direct connected devices
for 20% of the world’s current data generation. to do so.

A recent IDC report estimated that all edge devices—many

of them IoT ones—account for 20% of the world’s current
data generation.
 MIT Technology Review Insights 3

“Very small language

models now have the
sophistication to make
these machines more
intelligent. With hundreds
of words versus just a
single wake word, you
can start doing more to
control the objects
around you.”
Randy Abrams, Asia Pacific
Regional Technology Analyst, UBS

Today’s intense interest in IoT arises from various Accordingly, a whole technological subfield, sometimes
mutually reinforcing innovations within several of its called tiny machine learning (TinyML), has been
layers. Collectively, these advances will drive more emerging to allow such devices to use machine learning
storage, computing, and potentially decision-making and, increasingly, generative AI (hence the acronyms
capacity to the devices themselves—or the IoT’s AIoT and GenAIoT). Trade-offs inevitably exist: the
physical edge. This, in turn, could dramatically increase software on the edge may require extensive parameter
the utility of IoT technologies. pruning or approximations and, as a result, be less
accurate or have fewer capabilities. Nevertheless,
The highest-profile software development of recent the increase in user capacity from earlier generations
years has been the mainstream advent of generative of devices can be huge. Randy Abrams, Asia Pacific
AI, which has also brought renewed interest in older, Regional Technology analyst at UBS, explains that “very
predictive AI. Both rely on extensive data for training small language models now have the sophistication
or to populate their learning models. The volume of to make them [these machines] more intelligent. With
this data, however, can impede the ability of smaller hundreds of words versus just a single wake word, you
devices—including IoT—to use such software. can start doing more to control the objects around you.”

The highest-profile software development of recent years has

been the mainstream advent of generative AI, which has also
brought renewed interest in older, predictive AI.
4  MIT Technology Review Insights

Worldwide 5G mid-band coverage

5G mid-band population coverage was estimated to have reached approximately 40% worldwide by the end
of 2023.

World

World (without China)

North America

Latin America

Europe (without Russia)

Middle East and Africa

Asia-Pacific (without China)

China

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

5G total 5G mid-band
Source: Compiled by MIT Technology Review Insights, based on data from “5G network coverage outlook,” Ericsson, 2023

Meanwhile, the hardware to run this software is new standards to support IoT connectivity, is meant,
improving rapidly. Microprocessors, in particular, in part, to reduce this problem. So too will the greater
continue their decades-long increase in computational adoption of WiFi 6 and 6E, which are designed to allow
power. One result is a growing market for such chips faster local hub communication with a high number of
specifically designed to support AI on IoT and other low-power IoT devices. New specifications for WiFi 7,
edge devices. Competing companies include big released in January 2024, take these advances further.
names such as Qualcomm, Nvidia, and ARM. Already, Such innovations, says Danny Goh, founder and
Abrams notes, it costs only “a few dollars per device CEO of Nexus Frontier Tech, an AI research firm, are
to add small blocks of processing memory, and a bit “breaking the barriers to deploy smart home or other
of software development.” More advanced capabilities consumer devices freely anywhere you like.”
would increase the relative expense, but given
historical trends with information technology products, These mutually supportive developments will undergird
experts believe these costs could drop over time. the increasing integration of generative AI with IoT. Gartner
forecasts that, between 2023 and 2027, the global IoT
Although the above developments allow more market will grow at an average annual compound rate
processing and decision-making by IoT edge devices, of 14% and reach $124 billion. During that time, UBS
certain more complex tasks will inevitably continue estimates that the value of AI devices will grow from 4%
to require interaction between these assets and of the market to 12%. Meanwhile, in the same period,
centralized computing hubs. An important drawback consultancy GlobalData predicts that IoT software will
is the communication time required—or latency. The surpass IoT hardware in market size, consistent with
ongoing rollout of 5G mobile telephony, which includes greater computing on these distributed devices.
 MIT Technology Review Insights 5

GlobalData also projects that 27% of the total market Nor will evolution be restricted to traditional smart
will be for consumer IoT. It is therefore worthwhile to devices. Kitchen appliances are already also gaining
examine the likely effects of generative AI on consumer new features. For example, Whirlpool, Bosch, and
goods at the IoT’s edge and the key challenges to Samsung have introduced generative AI–enabled smart
implementing the technology. ovens, which can recognize a dish and then cook it
until they sense that the food is ready.
Innovation of two sorts
The impacts of generative AI on consumer goods As Abrams noted above, the cost of adding
fall into two broad categories: transforming already appropriate microprocessors to hardware upgrades
familiar products and allowing the creation of a whole as consumers replace old ones will be small compared
range of new ones. to the overall price of these tools. Goh adds that, in
many cases, “right now the sensors and all those
Abrams explains that it will be relatively straightforward smart devices are already very advanced.” The holdup
to upgrade any number of commonly owned household is “more on the software side,” he explains. However,
and personal smart tools, ranging from smart watches improvements could conceptually be distributed online.
and wearables, through vacuum cleaners and lawn
mowers, to smart thermostats and speakers. Goh Generative AI will also allow development of a wide
agrees. While the introduction of generative AI–enabled variety of new devices. Already, various existing or
devices into smart homes “is still very new at the soon-to-be-launched products combine in different
moment,” he adds, “things are starting to come in.” ways the newly available capacity of edge devices to
see, and act on, the world around them.
The addition of AI will give these devices substantial
new capabilities. Goh notes that even the restricted AI-enabled smart glasses with built-in cameras, such
versions of AI on edge devices will enable more as the Ray-Ban Meta Wayfarer and its competitors,
complex interaction and commands. The ability to are already available. The wearer of these devices
process and store data locally also presents new can request descriptions of images captured by the
possibilities. Goh believes that the resultant reduced camera, translate any text it sees, or ask for enriched
latency will be crucial for everything from making home information, such as where to buy a product seen via
vacuum cleaners respond to their environments more the glasses.
quickly to allowing use of facial recognition for home
security functions, such as unlocking doors. Moreover,
AI-based power management will extend battery
power and increase the physical range in which these
machines can be deployed.

“Right now the sensors and all those smart

devices are already very advanced. The
holdup is more on the software side.”
Randy Abrams
Asia Pacific Regional
Technology Analyst, UBS
6  MIT Technology Review Insights

These are only the beginning. Abrams reports that Barriers to, and risks arising from,
consumer robotics—currently a niche area— could
generative AI–enabled IoT
grow quickly depending on what the market provides.
While holding out great promise for consumers, a
Companies are certainly interested. Amazon has begun
variety of challenges could dramatically slow adoption
to sell a range of Enabot robots, which are marketed
or create unintended negative consequences from IoT
as family and pet companions as well as security
devices using generative AI. The following four are key
robots. Samsung will soon release its Ballie robot.
issues.
Both can act as smart device hubs and mobile voice
assistants to carry out various tasks. For example, they
Cost and an exacerbated digital divide. Expenses
can launch other IoT-connected equipment, monitor
related to integrating generative AI with IoT are
activity in the house for security and health issues,
as multifaceted as IoT itself. As mentioned earlier,
project videos, check emails, or even, when told, feed
additional microprocessors for specific devices
and play with the family dog.
will add relatively little to the price of already smart
devices. New machines will set buyers back more:
As ever with a new technology, the extent of innovation
smart glasses are currently in the low hundreds of
will reflect a combination of human imagination and
US dollars; some high-end robots can cost over a
consumer interest. Goh says that “things we only
thousand. The market will decide what is viable.
imagined just two or three years ago now can be a
lot more real.” Indeed, aspects of fairy tales may be
Changes to other IoT layers can be more expensive.
possible. A modern Sleeping Beauty cannot yet ask a
At the consumer level, Abrams explains that many will
mirror who is fairest of them all, but Baracoda’s smart
need to upgrade “to a much better home gateway that
mirror, BMind, will give mental health coaching to those
can deal with interference from a lot more signals and
looking into it.
a lot more devices within your home.”

IoT component interactions

The three components that make up IoT interactions are things with networked sensors and actuators
(TNSA), raw information and processed data storage (RI-PD-S), and analytical and computing engines (ACE).

Things with Networked Sensors Raw Information and Processed

and Actuators Data Storage
Internal and external
report states

Data stored

TNSA RI-PD-S

Analytical and Computing Engines

Feedback in the form Modeling and analysis
of commands of the data

ACE
Source: Compiled by MIT Technological Review Insights, “The Internet of Things: Insights into the building blocks, component interactions, and architecture layers,”
ScienceDirect, 2018
 MIT Technology Review Insights 7

The biggest expense, however, needs to go into on the IoT?” This novelty brings more than technical
underlying infrastructure. The rollout of 5G telephony, challenges. Abrams adds that companies will need to
let alone any eventual adoption of 6G, is expected to find application developers to create “a lot of different
provide substantial economic benefits, according things that are used in the home.” Given that this kind
to consultancy PwC. It will, however, require huge of generative AI is so new, that talent may be difficult,
investments. In the U.K. alone, the upgrade to 5G— and expensive, to secure, at least initially.
excluding rural coverage—is estimated to require £34
billion pounds by 2030, most of which will need to Lack of common standards. Consumers will want
come from state investment. This helps explain why, their IoT devices, and supporting layers, to work
while in the US and China over 90% of the population together seamlessly. This could take some time
have access to some version of 5G telephony, in to achieve. Currently, notes Goh, “the adoption of
Latin America, the Middle East, and Africa, the average [communication and other] standards in IoT is very
is 10%, according to Ericsson. different from provider to provider. Everyone is trying to
introduce their own.” Esposito adds that geopolitics will
On an even more basic need, access to electricity further slow standards development, citing the conflict
remains a challenge in many developing countries. between the US and China in establishing technical
Formal electrification data does not include reliability— norms for 5G. Companies, he adds, “cannot simply
this is clearly illustrated by data from health care, which innovate with the sake of scaling this out right. There
should presumably have privileged access to power. will always be some form of a geopolitical implication,”
According to a World Health Organization report, a including whether any advances raise perceived
billion people depend on health facilities in locations national security threats.
where the electricity supply is unreliable. They are
unlikely to have what they need for extensive IoT use. Safety, data privacy, and regulation. Deploying a
technology that can at times be unpredictable in
As Mark Esposito, a faculty affiliate at Harvard homes, and giving it access to substantial amounts
University’s Center for International Development, puts of personal information, brings risks unfamiliar with
it, globally the “infrastructure CAPEX requirements [for traditional consumer goods.
generative AI–enabled IoT] are gigantic. I am skeptical
that the pace of development will be on par with what Among these is the potential malfunctioning of
private firms are proposing.” automated devices. Emerging best practice with
generative AI tools in corporate use is to keep a so-
Still immature technology. Generative AI is a called “human in the loop,” who can stop processes
challenging technology to work with. As Abrams and equipment when they behave incorrectly. Abrams
points out, with its propensity for hallucinations, “it can foresees that consumer IoT devices will need to have
get things wrong and now it’s in your house helping not just automated shutdown but controls where
you.” Moreover, installations of the technology on people can do so both on site and via cloud-based
IoT devices will rely on stripped-down versions of applications. Esposito agrees that full autonomy for
learning models. Esposito points out that using these consumer IoT devices will not take place, if for no other
“is still largely in a testing ground. To what degree will reason than it remains unclear who might be legally
a [a smaller model] be functional to a specific use liable for mistakes.

“Globally the infrastructure CAPEX requirements for generative

AI–enabled IoT are gigantic. I am skeptical that the pace of
development will be on par with what private firms are proposing.”
Mark Esposito, Faculty Affiliate, Harvard University Center for International Development
8  MIT Technology Review Insights

“We now have more options to allow us to set up the security

barrier in each of the devices rather than transmit the data to a
central server.”
Danny Goh, Founder and CEO, Nexus Frontier Tech

Similarly, personal data leaks could occur in generative IoT devices, “simply put, we don’t have this yet,” says
AI-enabled IoT devices. Here, the technology itself Esposito. “Many governments are still trying to figure
can help. Goh points out that the capacity to use out the governance.” Goh agrees that key questions
more data on the device means that “we now have remain unanswered. For example, should legal liability
more options to allow us to set up the security barrier lie in the user, manufacturer, or software developer
in each of the devices rather than transmit the data if something goes wrong? Moreover, Esposito notes,
to a central server.” That said, much will depend on case law precedent on which companies could
implementation, especially as generative AI may bring potentially act, or officials base regulation, are also
with it new threat vectors for hackers to exploit. lacking. This will hold back progress more than
technological readiness. “I really hope that regulators
Ultimately, regulation will decide how these issues can give clearer guidance,” concludes Goh.
are addressed. Unfortunately, this remains nascent.
Regulation of generative AI as a whole was put in In short, while technology is moving rapidly to
place in the European Union and China in 2023, but in revolutionize consumer technology, any number of
many other jurisdictions, such as the US and Japan, factors mean that, in Esposito’s words, “this technology
it is either at the draft stage or non-binding. As for likely is not going to go as fast as the potential of what
regulatory specifics on the use of generative AI with it could do.”
 MIT Technology Review Insights 9

“Housetraining robot dogs: How generative AI might change consumer IoT” is an executive briefing paper by MIT
Technology Review Insights. We would like to thank all participants as well as the sponsor, UBS. MIT Technology Review
Insights has collected and reported on all findings contained in this paper independently, regardless of participation or
sponsorship. KweeChuan Yeo was the editor of this report, and Nicola Crepaldi was the publisher.

About MIT Technology Review Insights

MIT Technology Review Insights is the custom publishing division of MIT Technology Review, the world’s
longest-running technology magazine, backed by the world’s foremost technology institution—producing
live events and research on the leading technology and business challenges of the day. Insights conducts
qualitative and quantitative research and analysis in the US and abroad and publishes a wide variety of content,
including articles, reports, infographics, videos, and podcasts. And through its growing MIT Technology Review
Global Insights Panel, Insights has unparalleled access to senior-level executives, innovators, and
entrepreneurs worldwide for surveys and in-depth interviews.

From the sponsor

UBS is a leading and truly global wealth manager and the leading universal bank in Switzerland. It also provides
diversified asset management solutions and focused investment banking capabilities. With the acquisition of
Credit Suisse, UBS manages 5.5 trillion dollars of invested assets as per second quarter 2023. UBS helps
clients achieve their financial goals through personalized advice, solutions and products. Headquartered in
Zurich, Switzerland, the firm is operating in more than 50 countries around the globe. UBS Group shares are
listed on the SIX Swiss Exchange and the New York Stock Exchange (NYSE).

Illustrations
Illustrations assembled by Tim Huxford with elements from Shutterstock and Adobe Stock.

While every effort has been taken to verify the accuracy of this information, MIT Technology Review Insights cannot accept any responsibility or liability for reliance by any person
on this report or any of the information, opinions, or conclusions set out in this report.

© Copyright MIT Technology Review Insights, 2024. All rights reserved.

MIT Technology Review Insights

www.technologyreview.com
[email protected]