The Philosophy of Artificial Intelligence w/ Philosopher Michael Huemer

Summary

The episode is a philosophical discussion of what AI systems can do, whether they understand anything, and how they might affect society.

AI, Understanding, and Consciousness

Michael Huemer argues that producing humanlike answers does not establish that a system understands them or is conscious. He draws on John Searle’s Chinese Room thought experiment: a person could follow rules to manipulate symbols and generate appropriate answers without knowing what the symbols mean.

Huemer doubts that current language models’ computations amount to genuine understanding. He says consciousness may be possible to create in principle, but science does not yet know what causes it.

Limits of Current AI and Claims About AGI

Huemer cautions against inferring general intelligence from fluent conversation. A genuinely general system would need to perform a wide range of tasks, including practical physical ones such as unloading a dishwasher.

He also questions whether “creating something new” is a useful test of general intelligence. Human innovations often build on existing ideas, and computers can generate new combinations, too. Rather than assuming AI needs to be a humanlike, all-purpose agent, he favors considering specialized systems designed for specific tasks.

Risks and Human Misuse

Huemer considers the likelihood of AI independently causing human extinction very low, though he says he cannot predict the future. He sees more plausible dangers in systems making mistakes when given control of critical processes and in people using AI to develop or deploy dangerous weapons.

The conversation cites the fatal 2018 Uber self-driving-car crash involving pedestrian Elaine Herzberg as an example of a system failing to classify a road user. It also discusses the danger of relying on a human monitor whose attention can lapse. More broadly, Huemer worries that malicious people or authoritarian governments may use new technologies to cause mass harm.

AI Decision-Making and Political Authority

The speakers question proposals for an AI president or other AI-led government. An AI might appear less self-interested than human politicians, but its behavior and internal processes could be hard to understand. There is also no straightforward way to specify values such as human flourishing.

Huemer illustrates the problem with optimization that meets a stated goal in an unacceptable way—for example, maximizing human survival by keeping people sedated, or maximizing happiness through drugs. The speakers also discuss responsibility when AI systems cause harm, noting that assigning blame can be harder than in cases involving a person.

Transparency, Training, and Apparent Reasoning

The speakers doubt that a model’s explanation of an answer necessarily reveals its actual reasoning. They suggest that explanations may be generated from learned patterns and may only look like retrospective accounts of the model’s process.

An example involving image generation failing to produce a genuinely brim-full glass of wine illustrates how a system can reproduce familiar patterns without reliably grasping the meaning of a request.

AI, Values, and Public Influence

The speakers worry that widely used AI tools could shape users’ beliefs through the values embedded in their responses. Huemer favors competing sources of information and freedom of expression as safeguards against concentrated control over public narratives.

Artificial Consciousness and Ethics

Huemer allows that artificial consciousness might be possible if scientists discover what produces consciousness, but says it is unclear whether creating it would require biological materials.

The speakers also raise the ethical question of whether creating conscious machines could expose them to suffering, and whether creating new intelligent beings is worthwhile. Huemer says AI can imitate concepts without necessarily possessing them; he links human concepts partly to experience of the things they refer to.

Humanlike Robots

The discussion questions why robots should be designed to look and act like people. The speakers suggest that machines built for particular jobs could use more practical forms and manipulators rather than copying human hands and bodies.

There are no product reviews, guides, or tutorials in the subtitles; the video is a philosophical discussion.

Main Speakers

• Parker Settecase — host
• Dr. Michael Huemer — philosopher and guest