Gary Marcus expresses skepticism about the current state of artificial intelligence (AI) and its prospects for achieving human-like thinking. He highlights several key points:
1. Current Limitations: Marcus emphasizes that despite significant advancements in narrow AI applications, such as chess-playing or text translation, there has been limited progress in achieving strong AI that can exhibit human-level thinking and understanding.
2. Linear Progress: He notes that AI progress has been largely linear over the past five decades, despite exponential increases in computational power and memory capacity. Marcus suggests that the current technology is still far from achieving true AI.
3. Three Possibilities: Marcus presents three possibilities for achieving AI:
– Option (1): Solving AI with increased CPU power. He questions whether more computational power alone will lead to true intelligence.
– Option (2): Solving AI with better learning algorithms and larger datasets. Marcus points out that while this has led to advancements in specific tasks, it hasn’t resulted in machines with true thinking capabilities.
– Option (3): Solving AI by understanding the innate constraints or priors that evolution endowed humans with. He believes that understanding these priors is crucial for achieving AI that can think.
4. Bet on Option (3): Marcus places his bet on option (3), suggesting that understanding the innate constraints and priors that shape human thinking is the key to achieving AI with true thinking capabilities. He argues that the current focus on big data and computational power alone may not lead to genuine AI.
In summary, Gary Marcus expresses skepticism about the rapid achievement of human-like thinking in AI and emphasizes the importance of understanding the fundamental constraints that shape human cognition for future progress in the field.