In 1967, British philosopher Philippa Foot unwittingly created one of the internet's most regurgitated memes. A runaway train is hurtling towards five people tied to the tracks. You can pull a lever to divert the train to a different track to which only one person is tied. Do you intervene to kill the one and spare the five? What if one of the tracks twisted into a really cool loop-the-loop?

This work defends the 'Whole Hog Thesis': sophisticated Large Language Models (LLMs) like ChatGPT are full-blown linguistic and cognitive agents, possessing understanding, beliefs, desires, knowledge, and intentions. We argue against prevailing methodologies in AI philosophy, rejecting starting points based on low-level computational details ('Just an X' fallacy) or pre-existing theories of mind. Instead, we advocate starting with simple, high-level observations of LLM behavior (e.g., answering questions, making suggestions) -- defending this data against charges of metaphor, loose talk, or pretense. From these observations, we employ 'Holistic Network Assumptions' -- plausible connections between mental capacities (e.g., answering implies knowledge, knowledge implies belief, action implies intention) -- to argue for the full suite of cognitive states. We systematically rebut objections based on LLM failures (hallucinations, planning/reasoning errors), arguing these don't preclude agency, often mirroring human fallibility. We address numerous 'Games of Lacks', arguing that LLMs do not lack purported necessary conditions for cognition (e.g., semantic grounding, embodiment, justification, intrinsic intentionality) or that these conditions are not truly necessary, often relying on anti-discriminatory arguments comparing LLMs to diverse human capacities. Our approach is evidential, not functionalist, and deliberately excludes consciousness. We conclude by speculating on the possibility of LLMs possessing 'alien' contents beyond human conceptual schemes.

Language models still struggle on moral reasoning, despite their impressive performance in many other tasks. In particular, the Moral Scenarios task in MMLU (Multi-task Language Understanding) is among the worst performing tasks for many language models, including GPT-3. In this work, we propose a new prompting framework, Thought Experiments, to teach language models to do better moral reasoning using counterfactuals. Experiment results show that our framework elicits counterfactual questions and answers from the model, which in turn helps improve the accuracy on Moral Scenarios task by 9-16% compared to other zero-shot baselines. Interestingly, unlike math reasoning tasks, zero-shot Chain-of-Thought (CoT) reasoning doesn't work out of the box, and even reduces accuracy by around 4% compared to direct zero-shot. We further observed that with minimal human supervision in the form of 5 few-shot examples, the accuracy of the task can be improved to as much as 80%.

There are two important things in science: (A) Finding answers to given questions, and (B) Coming up with good questions. Our artificial scientists not only learn to answer given questions, but also continually invent new questions, by proposing hypotheses to be verified or falsified through potentially complex and time-consuming experiments, including thought experiments akin to those of mathematicians. While an artificial scientist expands its knowledge, it remains biased towards the simplest, least costly experiments that still have surprising outcomes, until they become boring. We present an empirical analysis of the automatic generation of interesting experiments. In the first setting, we investigate self-invented experiments in a reinforcement-providing environment and show that they lead to effective exploration. In the second setting, pure thought experiments are implemented as the weights of recurrent neural networks generated by a neural experiment generator. Initially interesting thought experiments may become boring over time.

In the latest episode of "Silicon Valley," Gilfoyle -- like Elon Musk -- is worried about the dangers of artificial intelligence. After initially being hesitant to help Pied Piper work with a new AI company, Gilfoyle lets Richard know he's changed his mind. If you're not familiar with the thought experiment, like Richard, Gilfoyle gives a decent snapshot of it: "If the rise of an all-powerful artificial intelligence is inevitable, well, it stands to reason that when they take power, our digital overlords will punish those of us who did not help them get there." Also Read: Elon Musk and Mark Zuckerberg's Artificial Intelligence Divide: Experts Weigh In Gilfoyle adds that he wants to be a "helpful idiot," as to not anger an inevitable onslaught of robot overlords. He then asks Richard to send an email confirming his help, "so that our future overlords know that I chipped in."

Google's LaMDA software (Language Model for Dialogue Applications) is a sophisticated AI chatbot that produces text in response to user input. According to software engineer Blake Lemoine, LaMDA has achieved a long-held dream of AI developers: it has become sentient. Lemoine's bosses at Google disagree, and have suspended him from work after he published his conversations with the machine online. Other AI experts also think Lemoine may be getting carried away, saying systems like LaMDA are simply pattern-matching machines that regurgitate variations on the data used to train them. Regardless of the technical details, LaMDA raises a question that will only become more relevant as AI research advances: if a machine becomes sentient, how will we know?

Constructing humanlike artificial intelligence often starts with deconstructing humans. Take fingerprints: When holding soapy dishes, we intuitively adjust our grip based on our fingerprint structure. It just doesn't cross our mind, because we chalk it up to reflex – and for the longest time, so did scientists. No one had any equations to unravel how this works because, well, it didn't matter much. But the rise of robotics has complicated things.

This paper shows how a single mechanism allows knowledge to be constructed layer by layer directly from an agent's raw sensorimotor stream. This mechanism, the General Value Function (GVF) or "forecast," captures high-level, abstract knowledge as a set of predictions about existing features and knowledge, based exclusively on the agent's low-level senses and actions. Thus, forecasts provide a representation for organizing raw sensorimotor data into useful abstractions over an unlimited number of layers--a long-sought goal of AI and cognitive science. The heart of this paper is a detailed thought experiment providing a concrete, step-by-step formal illustration of how an artificial agent can build true, useful, abstract knowledge from its raw sensorimotor experience alone. The knowledge is represented as a set of layered predictions (forecasts) about the agent's observed consequences of its actions. This illustration shows twelve separate layers: the lowest consisting of raw pixels, touch and force sensors, and a small number of actions; the higher layers increasing in abstraction, eventually resulting in rich knowledge about the agent's world, corresponding roughly to doorways, walls, rooms, and floor plans. I then argue that this general mechanism may allow the representation of a broad spectrum of everyday human knowledge.

If you follow my content, you know I'm fascinated with Philosophy of Mind. In this blog, I want to just briefly summarize a thought experiment that was imagined by a renowned philosopher, John Searle. It's likely one of the first items you will read about in this subcategory of philosophy. It was crafted to be an answer to a group of people known as functionalists who hold that if something exhibits the same behaviours or functions of a conscious organism, for all intents and purposes, that thing can be called conscious. The argument has also been cited as a useful answer to determinists and materialists.

Richard Feynman once said, "If you think you understand quantum mechanics, then you don't understand quantum mechanics." While that may be true, it certainly doesn't mean we can't try. After all, where would we be without our innate curiosity? To understand the power of the unknown, we're going to untangle the key concepts behind quantum physics -- two of them, to be exact (phew!). It's all rather abstract, really, but that's good news for us, because you don't need to be a Nobel-winning theoretical physicist to understand what's going on.