Well, huh. When I disagree with Peter Norvig, I'm usually wrong. No matter what problems I have with this claim that AGI is solved now (correction, not solved, has arrived, see comments) since we have LLMs, this article is well worth reading and thinking about since it is by one of the most respected people in the field, a guru of AI.

The dumbing down of #AI continues. We know for certain that #LLMs have no understanding (of causality, of truth, of the real world) and no ability to reason (they are happy to output X and not X in the same paragraph). Yet here is someone claiming that they somehow count as #AGI, as human-like intelligence. What, then, is intelligence? If it is truly possessed by LLMs today, then it is meaningless, and any computer program is suddenly promoted to be AI.

It's a bit difficult to say with certainty we have AGI, given that there is no precise consensus definition of it, but agree wholeheartedly that we should be taking a functionalist approach. Much to my surprise, for all practical intents and purposes, current AI systems have achieved the kinds of capabilities once depicted by e.g. HAL 9000 in movies like 2001: A Space Odyssey. Yet some people still argue that it's some kind of fancy billion-dollar parlour trick. Imagine a Neolithic human encountering a rocket for the first time. This person might dispute, from a theoretical standpoint, that a rocket, lacking wings, doesn't fit their concept of "genuine" flight. But this perception is limited by the observer's lack of understanding of physics. We now have sufficient understanding of physics that we can explain in great detail why e.g. a rocket can fly to the moon whereas a bird cannot. But we are very far from having a complete understanding of how minds work. Therefore, we should base claims about AI capability on empirical tests of what they actually can and cannot do, not on prescriptive theoretical ideas on how we think they should be doing it. People often point the dangers of anthropomorphizing LLMs, but imposing on them our own preconceived ideas about thinking, when those ideas have not led to well-developed and well-tested theories, is itself a form of anthropomorphism . At some point we will be able to explain exactly how LLM cognition differs from our own, but that will have to wait until we have better theories of cognition (both human, and artificial).

ICYMI: Check out last week's episode of Fed Gov Today to hear Finch AI Board Member Teresa Carlson address the importance of ATO for #AI, including how it can accelerate AI deployment for mission critical systems. #artificialintelligence #nationalsecurity

On this weeks episode of Fed Gov Today, Bridget Carper highlighted the significance of partnerships that value collaboration in testing the feasibility of implementing Artificial Intelligence (Technology) in government operations. She also emphasized the importance of architects (People) who possess a deep understanding of security, privacy, and data management, ensuring that these aspects can coexist without posing risks to organizations lacking robust cyber defense mechanisms. Unfortunately, without the correct partnerships, we cannot leverage collective expertise, resources, and perspectives to ensure successful and responsible AI deployment or mitigate risks while driving innovation. Here are a few more examples of how forming partnerships can positively impact Aritificial Intelligence in not only government operations, but also help shape meaningful policy on AI. 1. Access to expertise: Collaboration allows governments to tap into the expertise and knowledge of various stakeholders, such as industry experts, academic institutions, and technology companies. This collective expertise can help develop effective AI strategies, identify potential risks, and ensure the successful implementation of AI solutions. 2. Resource sharing: Partnerships enable the sharing of resources, including funding, infrastructure, and data. By pooling resources, governments can overcome budgetary constraints and leverage existing infrastructure to accelerate the adoption of AI in various government operations. 3. Holistic approach: Collaboration fosters a holistic approach to AI implementation. Different partners bring diverse perspectives and insights, ensuring that all aspects of AI deployment are considered, including ethical considerations, legal implications, and societal impact. This comprehensive approach should include stakeholders from state, local, tribal territories to make balanced, informed, decisions and address potential challenges effectively. 4. Mitigating risks: Implementing AI in government operations requires addressing various risks, such as data privacy, security, and bias. Collaborative partnerships allow for the identification and mitigation of these risks through collective problem-solving, sharing best practices, and developing inclusive frameworks and guidelines. 5. Accelerated innovation: Collaboration accelerates innovation by fostering knowledge exchange, experimentation, and the co-creation of solutions. Different partners can bring unique ideas and approaches to the table, sparking creativity and driving the development of cutting-edge AI applications that can benefit not only government operations, but also public services many of us rely on. #ResponsibleTech #partnerships #policy #policymaking #security

AGI is already here! AGI is not solved, but it has arrived. Like how ENIAC was the first general purpose computer in 1945 (but computing was not "solved") or like how powered flight arrived at Kitty Hawk in 1903 (but aviation was not "solved"). Read our article on Noema:

What do #space & AI have in common? As #AI hardware advances, innovators are developing AI tools capable of detecting hypersonic missiles in flight. Cool huh? And that’s just the start. More on AI hardware's transformative power in #ConnectingAI #39: https://lnkd.in/gwGjb5FC #artificialintelligence #machinelearning

“AGI is already here” — what’s Peter Norvig’s definition of #AGI? Let’s backtrack a little, what is intelligence? The latter is a philosophical debate and as we all know it, such debates rarely reach a universally accepted conclusion. If intelligence is undefined, it’s safe to say AGI has no definition. But rather than concerning ourselves with AGI, we should be dedicating ourselves to solving fundamental problems in AI like catastrophic forgetting so we can develop AI systems that are trustworthy & reliable.

Interesting thoughts from Peter Norvig

The more I use LLM tools, the more I'm convinced of this statement. Artificial General Intelligence is here but we're just seeing the very beginning of this new frontier today!