Gang Fang's Blog

Looking at state-of-the-art (SOTA) artificial intelligence (AI), which is based on deep learning (DL), through the lens of dual process theory, it’s easy to see that what SOTA AI truly accomplishes is the intuitive aspect of intelligence (in the broadest sense). This is akin to our ability to automatically coordinate our bodies when we run or produce language when we speak. According to dual process theory, the hallmark of skill expertise is the absence of conscious thought during its execution. In other words, the skill is performed intuitively.

It’s crucial to note that intuition involves two types of knowledge or expertise: one acquired naturally, either biologically or developmentally, and the other obtained through deliberate study or practice. Both share the common characteristic of being subconscious, yet we are significantly less aware of the first type, as well as the process of acquiring it. This explains our failure with logic-based expert systems in achieving AI: there’s a substantial amount of implicit knowledge that we aren’t aware of and don’t know how to produce. As a result, we couldn’t encode this knowledge into an expert system. Therefore, conscious reasoning cannot replace intuition.

Despite this, the gap between SOTA AI and artificial general intelligence (AGI) is System 2 AI — AI capable of conscious reasoning. This may seem counter-intuitive. So, why didn’t conscious reasoning AI lead to the current advancements in AI, but will instead catalyze future progress, building upon what we’ve already achieved using intuitive AI? The value of human conscious reasoning resides in the process itself, rather than in the knowledge it has produced so far, which is our civilization. In other words, our goal is to artificially replicate the biological or neurological mechanisms that enable us to discover knowledge, scientifically, artistically and generally. Evolutionarily speaking, what Yoshua Bengio has been proposing aligns closely with the human approach: this reasoning AI will be grounded in the artificial intuition produced by current AI and build on top of it to achieve general reasoning.