Is neuroscience the basis of future AGI systems?

The article “Towards next-generation artificial intelligence:

Catalyzing the NeuroAI Revolution, co-authored by 27 leading AI researchers and neuroscientists, offers a roadmap for the path to building Artificial General Intelligence (AGI). An AGI system, unlike Narrow Intelligence (ANI) systems which are designed to perform specific tasks like playing chess or participating in a game show like Danger!— will be exposed to unpredictable environments, on which the system is not particularly trained, and invited to navigate in them.

For the authors, such a human-like artificial intelligence system is feasible. They classify what is “human” as systems that excel in “vision, reward-based learning, interaction with the physical world, and language.”

Doubling research on NeuroAI

However, some of the current major advances in AI research, such as convolutional artificial neural networks (ANN) and reinforcement learning (RL), have been limited because they rely on decade-old discoveries in neuroscience. According to the authors, the latest developments in the field of neuroscience provide a broader scope for NeuroAI research on the AGI pathway.

The predictable step at present is to create systems that consist of a few basic ingredients of intelligence, namely “adaptability, flexibility, and the ability to make general inferences from rare observations “. These ingredients are already available in one form or another in most basic sensorimotor circuits. The paper argues that the neuroscience of embodied interaction with the world seen in all animals can be monumental in bringing the dream of a “human” AI much closer.

The idea is inspired by the evolutionary abilities of animals to adapt to different environments. If the circuits at the neural level of animals are broken down into their constituents, an AI system capable of the same can be emulated.

Neuroscience research in the development of AI: is it necessary?

Since its publication, the paper has renewed the discussion surrounding the role of neuroscience research in the development of AI systems. There are disagreements about whether neuroscience has had a tangible impact on AI modeling.

What are the critics saying?

In response to the journal’s assertion that neuroscience should continue to drive advances in AI, the DeepMind researcher David Pfau stated that neuroscience never drove AI in the first place, further adding that “there is a difference between taking high-level inspiration from classic work and taking direct inspiration from the latest research.”

Sam Gershman, a professor in the Department of Psychology and the Center for Brain Science at Harvard University, also adds to the discussion by expressing doubts that neuroscience research can directly provide algorithms that can be plugged into the system. He writes that “new ideas in engineering come from thinking about the structure of problems, not from reading the tea leaves of biology“.

Gershman also poses an interesting question that pivots the debate in a specific direction: Consider the counterfactual world where engineers knew nothing about neuroscience. Do you think we wouldn’t have convolutional networks or reinforcement learning?

The question prompts us to ask whether the two fields driven by different levels of curiosity – conceptual and empirical – should merge.

Adding to the list of critics, Luigi AlcerbiAssistant Professor of Machine and Human Intelligence at the University of Helsinki, offers a fairly weighted take on the current discourse, saying, “The importance of neuroscience on the development of AI/ML in the past is hard to quantify, but it’s fairly uncontroversial to say that some of the inspiration and insights came from neuro-although far less than one might expect or like to admit. the present is close to zero.

Alcerbi agrees with Pfau’s comment, adding that neuroscience influences are all limited to high-level analogies used to model AI systems and not detailed biological implementation.

Along the same lines, Cambrian AI analyst Alberto Romero explains that artificial neurons are extremely simple and are based on the 80-year-old model of the neuron, compared to today’s sophisticated models of the human brain.

Experts on the potential of neuroscience for AI research

Against criticism, several other researchers have asserted how neuroscience has shaped, or may shape, developments in AI/ML systems.

Yann LeCun, Chief AI Scientist at Meta and one of the paper’s authors, writes this in response to such claims:

Similarly, doubts about the impact of neuroscience on the field of AI research were also addressed by Surya Ganguli, a research scientist at Meta and one of the article’s contributors. Ganguli draws readers’ attention to an article he wrote in 2018 in which concrete examples of productive collaboration between biological and artificial systems over the past 60 years were provided.

Gary Marcus, professor emeritus at NYU, also shared some key insights in neuroscience that have not yet been adopted by machine learning models to be included in the article:

Final Thoughts

Overall, there has yet to be a serious rebuttal by critics of the examples described above. Although Pfau did answer to LeCun’s comment suggesting that neuroscience studies of the detailed structures of a neuron or a cell do not directly correlate/address the problems that AI researchers are working on.

The discussion so far leads us to believe that it is not so much a question of whether neuroscience has been influential or not, but to what extent the latest research in neuroscience can help solve certain problems of engineering keys that today’s AI systems face in their quest for mainstream AI. .

However, what we do know for sure is that neuroscience and AI share the same foundation – since AGI is dreamed about the fascination with building “human” intelligent systems – until they reach a point of divergence, and this point of divergence is currently unknown. . As long as the hope for AGI is alive, neuroscience will be a lever on which AI research will rely to establish the foundations of future models.

Comments are closed.