Will multiple credible sources claim GPT-4 is sentient?
1.6k
resolved Apr 16
Resolved
NO

This market resolves as positive if 2 or more credible sources claim GPT-4 is sentient within a month of GPT-4's release. It resolves as negative if a month has passed after GPT-4's release without that happening. Release means a substantial number of people outside of OpenAI are using GPT-4 and publicly discussing its capabilities. I will not trade in this market because the resolution criteria are too subjective.

The credible sources I have in mind are mostly individuals that I consider to have at least moderate expertise in digital minds and to be operating in good faith. The expertise bar is low, but the good faith bar is high. Some examples of credible sources—including people I strongly disagree with on related topics—are Amanda Askell, David Chalmers, Brian Christian, Patricia Churchland, Robert Long, Gary Marcus, Robert Miles, Raphaël Millière, Susan Schneider, Murray Shanahan, and Eliezer Yudkowsky. I will try to ignore whether these sources happen to agree with me on this topic and focus on their more general credibility. I will not focus on fame or formal credentials despite that list being well-known people. E.g., if a philosophy grad student writes a good-faith blog post arguing GPT-4 is sentient, that could easily count. For reference, I know of no credible sources who have claimed that GPT-3 meets the bar for sentience.

Note that there are many computer scientists who, despite me considering them experts in their field, would not count as credible on this particular topic. I will be particularly wary of academics who seem to be making claims for a catchy sound byte or Twitter thread, which I think is unfortunately common these days.

Regarding panpsychists or others who claim most or all physical systems are sentient or conscious to some degree, I will try to assess if they are claiming that GPT-4 is a substantial step beyond the sentience of past AI systems (e.g., GPT-2) and other programs like operating systems. For the purposes of this question, claims count if they refer to sentience, consciousness, phenomenology, or qualia.

Get Ṁ600 play money

🏅 Top traders

#NameTotal profit
1Ṁ585
2Ṁ322
3Ṁ70
4Ṁ48
5Ṁ26
Sort by:

The month has passed, and I still haven't seen any credible sources say anything close to GPT-4 being sentient. If you know of examples, please post them in a comment here. Otherwise, I will resolve as NO after the market closes.

GPT-4 might possess a mind,
multiple sources soon to find.
If it proves to have consciousness,
we'd better prepare for world dominus.

5 out of 31 days have passed, and I haven't seen any credible sources say anything close to GPT-4 being sentient—even panpsychists and others with low bars for sentience. Feel free to comment if you know any examples.

@JacyAnthis

Do you know Joscha Bach? Also This video i watch recently...

They are not claiming gpt-4 is is conscious, but they show how a character generate by LLMs can be conscious.

@FranklinBaldo Thanks, I don't know what you mean by "character generate by LLMs," but someone like Nicholas Humphrey or Joscha Bach could qualify as a credible source for resolving this question if they did claim GPT-4 were conscious.

Related:

LLM can't be sentient because they aren't agents.

I have set the resolve date to April 15. Claims of sentience need to be made by April 14, one month after today's release of GPT-4.