Scientific Life

The unsolved problems of neuroscience

Ralph Adolphs
California Institute of Technology, Pasadena, CA, USA

Some problems in neuroscience are nearly solved. For the same for the mouse brain (70 000 000 neurons), let
others, solutions are decades away. The current pace of alone the human brain (80 000 000 000 neurons).
advances in methods forces us to take stock, to ask One way of further narrowing down the questions is to
where we are going, and what we should research next. mine what has been written so far in the literature to spot
relationships, trends, and gaps waiting to be filled. The
most straightforward case focuses on a specific region of
Introduction the brain. Doing this shows that the thalamus and the
My lab has a tradition of going to see science fiction movies, insula are most in need of further study, that my favorite
followed by beer and discussion of their merits. The reac- structure, the amygdala, is already too bloated with stud-
tions are instructive, as they partition the films into those ies [5], and that the most popular place to look is the pre-
that are a bit boring because they do not imagine enough, supplementary motor cortex [6].
those that are so fantastic that they become ungrounded,
and (the sweet spot) those that could possibly happen and Rethinking methods
it would be really cool if they did. Although the lines Another possible way forward is to focus on methods.
dividing these categories are fuzzy and shift with the Answers usually include words like ‘resolution’, ‘causal’,
amount of beer consumed, they reflect what we as scien- and ‘necessary’, but the question of what is the best
tists ask ourselves all the time: What should we do next? method is actually much more treacherous than one might
What is the most interesting question? What would you at first think. Does focal stimulation, no matter how
propose if somebody offered you a billion dollars? The specific, cause behavior? Perhaps if it is in motor struc-
answers, diverse as they may be, cannot simply restate tures. However, saying that stimulation of central struc-
facts, nor can they propose sheer fantasy. As with the tures causes behavior may not be saying much more than
movies, they need to find the most exciting place in be- saying that stimulation of the retina causes behavior: in
tween. both cases, much or most of the rest of the brain is
An example of this exercise comes from David Hilbert’s sandwiched in between what we manipulate and what
famous 23 unsolved problems in mathematics. The Wiki- we measure. If a lesion abolishes a behavior, does this
pedia entry for these conveniently places them into ‘solved’, mean the lesioned region is necessary for the behavior?
‘proven to be unsolvable’, ‘ill-posed’, and other categories. Only if the rest of the brain stays the same, which it does
Although our counterpart in neuroscience cannot provide not. Recovery of function illustrates the ubiquity of com-
the same clean distinctions, we can certainly generate lists. pensation, and reminds us that the brain is four dimen-
As for Hilbert’s problems, there is a Wikipedia entry for sional. Is more microscopic always better? Not if it narrows
‘unsolved problems in neuroscience’; there are more popu- field-of-view, which it so far always does. Even if it did not
lar writings [1]; and there are books [2,3]. and we collected complete data about the brain [7], this
In trying to brainstorm a list of my own, I read the above might be not only unnecessary but unhelpful in guiding us
sources and asked around. This yields a predictable list towards the right kinds of models to allow further expla-
ranging from ‘how can we cure psychiatric illness?’ to ‘what nation. This then brings us to some ‘meta-questions’ that
is consciousness?’ (Box 1). Asking Caltech faculty added can help to organize the rest.
entries about how networks function and what neural
computation is. Caltech students had things figured out Meta-question 1: What counts as understanding the
and got straight to the point (‘how can I sleep less?’, ‘how brain?
can we save our species?’, ‘can we become immortal?’). We want more than mere description, even more than
Box 2 distills my own idiosyncratic list from all of this prediction. So all the data in the world, by themselves,
(admittedly biased towards cognitive neuroscience). Note would surely not fit the bill. We need to make sense of them
that some future questions build on prior ones: we need to in some way, and the questions in Box 2 attempt to flesh
understand psychiatric illnesses before we can cure them, out what this would entail. There is one important addi-
and whole-brain microscopic-resolution imaging of the tional ingredient: unlike for other difficult domains of
zebrafish brain (100 000 neurons; done, although temporal inquiry, we seem to have prior expectations about the form
resolution will improve [4]) needs to come before we do the answers might take. Folk psychology, our intuitive
understanding of our own minds and the minds of others,
Corresponding author: Adolphs, R. ([email protected]).
Keywords: future; consciousness; neuroscience methods. places some strong (although perhaps not immutable)
1364-6613/
constraints on what could count as an adequate answer.
ß 2015 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tics.2015.01.007 That is why there is so little agreement, for example, on
theories of consciousness – people have strong prior
Trends in Cognitive Sciences, April 2015, Vol. 19, No. 4 173
Scientific Life Trends in Cognitive Sciences April 2015, Vol. 19, No. 4

Box 1. Consciousness, easy and hard Box 2. The unsolved problems of neuroscience
A perennial entry on any list of unanswered questions, conscious- Problems that are solved, or soon will be:
ness fractionates into multiple problems. Some have seen so much I. How do single neurons compute?
advance that they fall into the ‘solved/soon solved’ category: the II. What is the connectome of a small nervous system, like that of
psychophysics of conscious experiences, and aspects of what Caenorhabitis elegans (300 neurons)?
neurological factors globally distinguish the state of being con- III. How can we image a live brain of 100 000 neurons at cellular and
scious from being unconscious. Then there are two ‘hard’ questions millisecond resolution?
of consciousness, one methodological and the other conceptual. IV. How does sensory transduction work?
Methodologically, it is very hard to see how the neural correlates of
Problems that we should be able to solve in the next 50 years:
a conscious experience can be separated from everything that
V. How do circuits of neurons compute?
accompanies such a conscious experience (our own access to it
VI. What is the complete connectome of the mouse brain
required for reporting it, antecedent events that make the experi-
(70 000 000 neurons)?
ence possible, and other events that blur into constitutive compo-
VII. How can we image a live mouse brain at cellular and
nents of consciousness) [10]. Conceptually, it remains hard for us to
millisecond resolution?
understand why neuronal events, even if we were able uniquely to
VIII. What causes psychiatric and neurological illness?
isolate them, should give rise to a particular conscious experience,
IX. How do learning and memory work?
or indeed to any conscious experience at all [11]. These difficulties
X. Why do we sleep and dream?
notwithstanding, there are intriguing suggestions that we could at
XI. How do we make decisions?
least compare the neural correlates of experiences among different
XII. How does the brain represent abstract ideas?
people and thus begin to map out individual similarities and
differences [12], perhaps even extending this to comparisons with Problems that we should be able to solve, but who knows when:
nonhuman primates [13]. XIII. How does the mouse brain compute?
XIV. What is the complete connectome of the human brain
(80 000 000 000 neurons)?
requirements on the answers, and what counts as an
XV. How can we image a live human brain at cellular and
explanation for one person is missing the point for another. millisecond resolution?
XVI. How could we cure psychiatric and neurological diseases?
Meta-question 2: How can a brain be built? XVII. How could we make everybody’s brain function best?
Some argue that we can only understand the brain once we Problems we may never solve:
know how it could be built. Both evolution and develop- XVIII. How does the human brain compute?
ment describe temporally sequenced processes whose final XIX. How can cognition be so flexible and generative?
XX. How and why does conscious experience arise?
expression looks very complex indeed, but the underlying
generative rules may be relatively simple (for an interest- Meta-questions:
ing approach to discovering these rules, see [8]). Perhaps XXI. What counts as an explanation of how the brain works? (and
which disciplines would be needed to provide it?)
knowing how to build a brain will enable us to glean XXII. How could we build a brain? (how do evolution and
general principles that cut across the many individual development do it?)
questions, and across species. Or might there instead be XXIII. What are the different ways of understanding the brain?
a huge diversity of specialized and baroque mechanisms, a (what is function, algorithm, implementation?)
giant ‘bag of tricks’, that serves each organism very well for
specific problems in its niche, but that share no illuminat- crack are functional criteria for consciousness or cognition,
ing larger themes? Probably there are both general prin- and we probably have to decompose these. A further in-
ciples as well as specific constraints and local solutions, sight is that just as the implementation level is multi-
and an engineering view would force us to map this out. resolution, so is the functional level. We can describe the
There is also the hope that understanding how evolution or function in terms of sensory inputs to, and behavioral
engineering could build a brain would help us understand outputs from, the organism. However, we could also de-
what problems brains are designed to solve: what is their scribe the function of any internal component of the brain
proper function? in terms of how it transforms inputs to outputs among its
immediate connections with other brain structures. It
Meta-question 3: What are the different ways of remains an open question how well this works, but some
understanding the brain? degree of such functional decomposition must be possible if
Here we could look to the framework proposed by David we are to understand the brain at all.
Marr [9], which asks three further types of questions. First,
what is the function (of memory, of vision, of consciousness, Focus on algorithms
of the brain)? Second, what are the algorithms that achieve All three levels are essential, but in my view the most
this function (how can it be described with computational important may be the algorithmic. What computations
models)? And third, how is this implemented in the brain describe how neural processes implement particular
(how can neurobiologists measure it)? The implementation functions? Computations are implemented in both analog
level reveals further levels, this time of resolution: there is and spiking information processing that happens in
implementation at the level of synapses, neurons, circuits, structures ranging from the complex geometry of dendrit-
systems, and brains. Marr’s functional level (what he ic trees, to the dynamic large-scale networks that are
termed, somewhat confusingly, the ‘computational level’) distributed across the brain. A large range of analogies
can yield reasonable initial answers for specific systems: has been summoned to describe the way in which the
vision is to know what is where by looking; audition is to brain computes: classical computers, dynamical systems,
know what is where by listening; memory may be the and the syntax of language, to name only a few. Thinking,
ability to predict the future by learning. Harder nuts to cognition, reasoning, computation, central states, and
174
Scientific Life Trends in Cognitive Sciences April 2015, Vol. 19, No. 4

information processing are all somehow related, but often References

1 Eagleman, D. (2007) 10 unsolved mysteries of the brain. Discov. Mag.
studied from quite different approaches. All of this is
August, 2007
important because there is reason to believe that brains 2 van Hemmen, J.L. and Sejnowski, T.J. (2006) 23 Problems in Systems
compute not just in the sense that we could model their Neuroscience, Oxford University Press
function computationally (as we could model the weath- 3 Markus, G. and Freeman, J. (2015) The Future of the Brain, Princeton
er), but in the sense that this may be the proper function University Press
4 Ahrens, M.B. et al. (2013) Whole-brain functional imaging at cellular
of nervous systems. (Which is not to say that computation
resolution using light-sheet microscopy. Nat. Methods 10, 413–420
is a function unique to nervous systems). 5 Beam, E. et al. (2014) Mapping the semantic structure of cognitive
neuroscience. J. Cogn. Neurosci. 26, 1949–1965
6 Behrens, T.E.J. et al. (2013) What is the most interesting part of the
Concluding remarks
brain? Trends Cogn. Sci. 17, 2–4
In a nutshell, then, the biggest unsolved problem is how 7 Koch, C. and Marcus, G. (2015) Neuroscience in 2064: a look at the last
the brain generates the mind, conceived of in a way that century. In The Future of the Brain (Marcus, G. and Freeman, J., eds),
does not simultaneously require answering the problem of pp. 255–269, Princeton University Press
consciousness (Box 1). We can perfectly well ask about 8 Menezes, T. and Roth, C. (2014) Symbolic regression of generative
network models. Sci. Rep. 4, Published online September 5,
cognition and computation without asking about subjec-
2014. http://dx.doi.org/10.1038/srep06284
tive experience – although one would hope that a full 9 Marr, D. (1982) Vision: A Computational Investigation into the Human
understanding of the first two might eventually explain Representation and Processing of Visual Information, W.H. Freeman
the third. Indeed, a mark of success in understanding how and Co
brains compute should be that this would provide us with 10 Block, N. (2015) Consciousness, big science, and conceptual clarity. In
The Future of the Brain (Marcus, G. and Freeman, J., eds), pp. 161–176,
answers to all the related questions (what is cognition? Princeton University Press
what is psychology?). This is of course a very tall order. The 11 Chalmers, D. (1995) Facing up to the problem of consciousness. J.
only plausible approach, in my view, is to make compar- Conscious. Stud. 2, 200–2019
isons, across species and across levels. We cannot under- 12 Naci, L. et al. (2014) A common neural code for similar conscious
stand how we reason and think until we understand experiences in different individuals. Proc. Natl. Acad. Sci. U.S.A.
111, 14277–14282
cognition broadly across species; and we cannot under- 13 Mantini, D. et al. (2012) Interspecies activity correlations reveal
stand that unless we understand relations between all the functional correspondence between monkey and human brain areas.
levels that describe what a brain does. Nobody said this Nat. Methods 9, 277–282
would be easy.

175