Home » Uncategorized

Can Neuroscience unlock the doors to superhuman AI?

Human brain is arguably the most complex material human kind will ever claim to fully understand. The complexity arises somewhat from the fact that there are many (different but compatible to each other) ways of talking about it. One way to talk about it is that it’s made of the same elementary particles as Martian sand dunes and obeys the same physical laws of nature governing violent collision of two massive black holes. Another way to talk about it would be that it’s guilty of two world wars but is credited for Great Pyramid of Giza and The Starry Night. It is somewhere in the middle of the two vernaculars that neuroscience has made considerable progress in recent years.

Fostering Neuroscience for accelerating Artificial Intelligence and Machine Learning – fool’s errand?

Neuroscience is a multidisciplinary branch of biology that focuses on the nervous system and attempts to describe various neural phenomena in teams of anatomy, biochemistry, molecular biology and physiology of neurons and neural circuits. With recent advances in sophistication of computer systems and its computational abilities, the industry now more than ever is claiming the inevitability of AI taking over much of human activities that at present keep hundreds of millions of us gainfully employed. If you have recently been pitched how deep learning is breaking ground during a water cooler chat, you are more likely than not sympathetic to your fellow conversationalist’s concern that AI is purposefully on its way to replace both of you out of your 9 to 5 grind. I am neither arguing for the likelihood of that scenario playing out in our lifetimes, nor trying to incite a debate on how we can be ready as a society to cope with existential void it might bring which is now mostly filled with sense of meaning and purpose trying to put food on the table. Rather the intent is to emphasize that if you are an AI developer working on an algorithm to dethrone the best of human gamer; you may be justified looking into behavior of neural networks or computational neuroscience for clues. This explains why the fields of neuroscience and AI have a long and intertwined history. However, more recently, interactions between the two disciplines have become less commonplace; as both subjects have evolved, growing enormously in complexity and solidifying disciplinary boundaries. I think the net outcome of this separation is positive for both fields. Let’s probe.

Why modern jetliners don’t have flapping wings?

Nature often gives us the best design solutions. Taking a little inspiration from termite architecture for a large building on a cheap or designing a speedy swimsuit after hydrodynamic properties of shark skin are examples of bio-mimicry at its best. However, copycatting from nature is an exercise whose de facto standard is trusting millions of years of evolution over a few decades of human research and development. This is problematic for mainly two reasons.

1.) Evolution is never-ending. It’s guided by natural selection through random mutations towards modification of gene pool of a population that makes the population more robust to its environment than its ancestors. In other words, human brains are still evolving.

2.) Evolutionary processes are not required to follow or produce best possible architectural solutions. Therefore many of our organs are arbitrarily limited in functionality. Why our eyes are arbitrarily limited to sense only a tiny sliver of the electromagnetic spectrum?

How AI beats humans at specific tasks?

The 1997 rematch of Deep Blue versus Garry Kasparov was the second contest of six-game chess matches between world chess champion Garry Kasparov and an IBM supercomputer called Deep Blue.

The 1996 match

Game #

White

Black

Result

Comment

1

Deep Blue

Kasparov

1–0

 

2

Kasparov

Deep Blue

1–0

 

3

Deep Blue

Kasparov

½–½

Draw by mutual agreement

4

Kasparov

Deep Blue

½–½

Draw by mutual agreement

5

Deep Blue

Kasparov

0–1

Kasparov offered a draw after the 23rd move

6

Kasparov

Deep Blue

1–0

 

Result: Kasparov–Deep Blue: 4–2

The 1997 rematch

Game #

White

Black

Result

Comment

1

Kasparov

Deep Blue

1–0

 

2

Deep Blue

Kasparov

1–0

 

3

Kasparov

Deep Blue

½–½

Draw by mutual agreement

4

Deep Blue

Kasparov

½–½

Draw by mutual agreement

5

Kasparov

Deep Blue

½–½

Draw by mutual agreement

6

Deep Blue

Kasparov

1–0

 

Result: Deep Blue–Kasparov: 3½–2½

Deep Blue’s win over the Kasparov in 1997 was symbolically significant. It signaled that AI was at the very least catching up to human intelligence in specific domains. The list of these specific domains is growing since. We now have machines that are better poker players than us. At the heart of Deep Blue’s technique was a purely mathematical construct (Evaluation Function) which assigns a real number to each possible states of board arrangement at any given time. As far as we’ve looked, this mechanism is nowhere to be found in nature even in its adulterated form. I would argue that this is compatible with what we know about the physical world so far. 

An anthropomorphic approach to the disciplines of AI is bound to restrict our expectations from them and can impede progress. Scientific inquiry that is inherent to all of us is enough of a reason to motivate us to further neuroscience as a discipline in the same manner it drives us to detect gravitational waves.

Leave a Reply

Your email address will not be published. Required fields are marked *