Video Gamer is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Prices subject to change. Learn more
For those wondering if a possible GPT-5 could have artificial general intelligence (henceforth AGI), don’t worry. You’ve come to the right place.
With the release of GPT-4 and multiple other AIs like Google Bard and Microsoft Copilot in the first half of 2023, the development of AI is a topic on many people’s minds. It is interesting that many of the AIs on the market are predominantly billed as assistants to human agents rather than more distinct self-reliant programmes (of the more sci-fi-esque super-computers mould that fiction has sensationalised for decades.) The reason for this can probably be laid at the feet of the AI/AGI differentiation – we’ll get into that, don’t worry. But, with all the advances we have seen already, could people reasonably expect a version of GPT-5 to have AGI?
Let’s have a look at what we think.
Jasper AI
Copy AI
Originality AI detector
Could GPT-5 have artificial general intelligence (AGI)?
First off, we should state that there has been no confirmation of a GPT-5, nor of a possible release date. However, we will be using the term GPT-5 in this article as a shorthand for the next possible model of ChatGPT that OpenAI may come out with following on from GPT-4.
Next, we need to explain a bit of some of the context surrounding this discussion. Let’s get into the first issue:
The AI/AGI divide
To summarise a complicated issue, this divide comes down to AI being a machine designed to be similar to humanistic thought processes but not necessarily capable of complete replication of capacity. AGI, by contrast, is a more complicated potentiality entirely, generally considered to be a much harder benchmark for AI developers to hit.
AGI has created a certain amount of difference of opinion since it’s conception. Common understandings of AGI include an AI that behave with a similar intelligence ability as a human. However, this is rather vague. Some have positied that AGI classification could also rest on such parameters as ‘learning and flexibility.’ This was certainly a theory that was posited around the time of 2019 by Henry Shevlin, Karina Vold, Matthew Crosby and Marta Halina in this paper on ‘The Limits of Human Intelligence’. Since then, however OpenAI has released ChatGPT and GPT-4 which both demonstrate flexibility and learning, to a certain degree, and yet have not generally received AGI classification.
On the other hand, there are some more concrete tests that could provide us with a clear divide between programmes deserving of AGI classification or not. For example the Winograd Schema Challenge, which was developed as an improvement on the Turing Test. However, GPT-4 reached almost human-level scores by with 87.5% (5-shot) in the WinoGrande test, (for reference human-scores were determined by the original test makers as 94%), however AGI status for GPT-4 is an issue that has no clear resolution at the time of writing (although people do say there might be sparks). Our conclusion on this one, as others have also come to, is that before AGI classification can be handed out with any kind of authority, a more clear cut definition would be needed.
This has been a very whistle-stop tour through a very complicated issue. For a more detailed investigation into the AI vs AGI issue please head over to our article on the matter.
It seems that there is more to the situation than what we first thought.
Could GPT-5 have AGI?
That being said, with the definitions we currently have of AGI, we can do some speculating about the possibility of a the possible next ChatGPT model having AGI. Here’s what we think.
Reasons why we think a GPT-5 probably won’t have AGI:
The last word from OpenAI on AGI potential we have to go off is in an interview with Lex Fridman, Sam Altman (CEO of OpenAI) posted on March 25, 2023. In the interview Altman claimed that there were barriers to AGI achievement with their technology that the team did not currently know how to overcome. He stated that they would need to ‘expand on the GPT paradigm in pretty important ways that we’re still missing ideas for.’ This makes it clear that OpenAI does not personally see how they are going to achieve AGI, as yet.
Further supporting that view was his next comments: ‘if an oracle told me far from the future, that GPT-10 turned out to be a true AGI somehow, you know, with maybe just some very small new ideas, I would be like, okay, I can believe that. Not what I would have expected sitting here, I would’ve said a new big idea, but I can believe that.’ The general sentiment of uncertainty surrounding how to unlock AGI potential seems clear. Altman appears open to either option of a large GPT overhaul being that unlocking stage, or if the solution lay within reach but, for now, out of sight.
The mention of ‘GPT-10’ is interesting, however. The clear sentiment of hope towards AGI potential within OpenAI technology is clear. However, while Altman appears hopeful for AGI capacity, he does not see it as a close enough prospect to attach GPT-5 timeframe possibility to. Instead, he pins the possibility in a distant future, multiple iterations down the line.
Reasons why it could be possible that a GPT-5 might have AGI:
In that same interview referenced above, Altman claimed that he considered AGI to be a system that could ‘significantly add to the sum total of scientific knowledge’, a system that could ‘discover, invent, whatever you want to call it.’ This would add creativity to one of the parameters of AGI. This is interesting as one of the key selling points highlighted by OpenAI for GPT-4 was was it’s ‘creativity.’ This is clearly something that is on OpenAI’s mind when it comes to the development of their AI. This could suggest that the next instalment of OpenAI’s GPT series could be reasonably expected to be a bid for AGI.
This view is supported by entrepreneur, Siqi Chen. Chen posted a tweet on March 27, 2023 stating ‘I have been told that GPT5 is scheduled to complete training this December that openAI expects it to achieve AGI.’ However, the reliability of the information contained within this tweet can be called into question, due to Chen’s lack of sources provided and it’s incongruence with Altman’s sentiment of a later model being the more likely to achieve AGI. That being said, Chen’s tweet was posted on March 27, 2023. While it is possible that within two days, Altman’s perspective changed, this has not been publicised.
Final Words
Many issues surrounding the discussion of potential AGI capacity in the next GPT iteration make predicting anything with any kind of reliability a difficult task. While others have certainly made the leap, with Shane Legg predicting ‘roughly human level AGI in the year 2025’ in 2009 on a vetta project blog post that you can find here, we find that kind of statement harder to make with certainty. In our opinion, one of the largest issues is defining what is actually meant by AGI capacity. One of the many issues that has been raised in this regard is that the term intelligence itself is inherently vague and open to interpretation. That being said, we do believe that AGI is a possibility, whether developed by OpenAI, or not. Indeed, others have felt it to be enough of a possibility that research has been undertaken as far back as 2014 into the mitigation of AGI development.
For all the gaming news you might need, make sure to watch this space. We’ve got you covered from
Frequently Asked Questions
Are there any AGIs out now?
This is a hotly debated topic. At the time of writing we cannot say that there are any confirmed AGIs available right now for public use, or indeed, in existence at all.
Is artificial general intelligence possible?
The CEO of OpenAI, Sam Altman, stated in an interview that, ‘if an oracle told me far from the future, that GPT10 turned out to be a true AGI somehow… I would be like, okay, I can believe that.’ However, this is a single opinion and there is much debate on the topic. As such, at the time of writing, a clear answer is difficult to provide.
