From recognition to cognition
The beginning of AI could be characterized as sophisticated pattern recognition. ML can identify patterns in data, such as classifying whether an image is of a dog or a cat or if a radiograph contains cancer. This is extremely valuable for a broad class of problems.
Building on this, once a pattern can be reliably identified, then a prediction can be made about what’s next. For example, based on a set of real estate sales data, a prediction can be made about the next home for sale.
Arguably, detecting cancer is of greater importance than home values. But in terms of comparing artificial intelligence to human intelligence, predicting the future is a higher order capability than pattern recognition.
Recently, we’ve seen an explosion of the next evolution in AI: generation of speech, images, video, and more. AI has been generating this type of content for some time, but over the summer, we saw the release of AI models that moved from okay to very good.
To follow the prior line of reasoning, the ability to predict what’s next is a necessary precursor to generating something new. A painting is a flow of color, strokes, and patterns (or pixels as the case may be) that require continuous “what’s next” decision-making.
If you agree with this progression: pattern recognition -> prediction -> generation, then we may speculate on how AI will evolve. The ability for AI to generate music, art, and novels are all precursors for bi-directional communication.
There’s a strong argument that the current state of AI is only piecing together a set of known concepts and presenting them as new. However, what AI can do is consistently surprise and delight. And if creativity is the act of creating something new, then AI meets that standard today.
But, what AI can’t do yet, is reason and communicate its rationale. AI lacks context and goals; self-awareness and self-direction. However, AI is evolving at an extremely rapid pace. And importantly, extensive research is under way on how to make neural networks recursive and to use AI to make AI.
Some have characterized the next step of evolution as the Third Wave of AI (or Cognitive AI): https://www.darpa.mil/attachments/AIFull.pdf. The suggestion is that AI will be able to contextually adapt to what’s it’s presented - to reason and respond.
Importantly, this step probably won’t be a full AGI with complete autonomy and its own open-ended goals, but it likely will feel very real to us in a broad set of situations.
As such, it’s reasonable to suggest that the next step of evolution for AI is Cognition.
Skeptical of AI moving to Cognition? Here’s Intel asserting it will happen in the next three years:
By 2025, it is expected that human-centric Cognitive AI systems with higher machine intelligence will emerge. Machines will be able to understand language, integrate commonsense knowledge and reasoning, and adapt to new circumstances. https://www.intel.com/content/www/us/en/research/blogs/higher-machine-intelligence-for-next-gen-ai.html