How long before Artificial Intelligence becomes your client or your adversary?

If you have not encountered ChatGPT in your daily news feed yet, you may want to be careful about who you are conversing with.
The latest iteration of the chatbot from the OpenAI team, released in November 2022, is now causing a storm and predictions ranging from Dystopia to Utopia. Kids (Just kids?) are using it to write their homework, musicians are getting upset about it writing songs, software developers are nervous as it can write code and it has been shortlisted for a job after writing its own application. Now, we are pondering how long it will be before it can deal with your divorce or next house purchase.

There is no doubt this technology is mind-blowing. Its sibling named Dall-E2, from the same OpenAI stable has been causing commotion for some time with the video, sound and images it creates. To get where it is now with version 3.5, ChatGPT was fed just 570GB of text. I say just, as although that is about 300 billion words, it is still an imperceivable fraction of what is available on the internet. It is also learning as the frenzy builds and wave after wave of people try it out in various guises.

A downside of the information it is learning from, is it is subject to the underlying bias of the humans who have written it. The developers have tried to take this into consideration and the responses given are subject to code that acts as a moderator. Which is fine as long as the humans writing it, believe they can outmanoeuvre a creation that has the objective of getting better. If you engage with ChatGPT it remembers the conversation and responses given so far. When it detects a question that is not factual it can overcome this dilemma by posing the response as a hypothetical scenario. This means you can ask it what happened when we landed on Mars and potentially get a work of fiction in return. However, it is also subject to “hallucinations”. Yes, a real term used in this context to describe when it gives incorrect or confused responses.

What will this do for the Law, given we have already seen it is close to passing the exam even in its current state? Well, just think about what makes a good lawyer? Is a really good lawyer one that has apparent limitless knowledge of legislation and case law? Somebody who has eidetic memory of all courtroom transcripts and can process the statements and responses in each case that produced a positive outcome? Somebody that can recall every word on your title deed and interpret it because it is from Cornwall and written in kernewek? Somebody that can (appear to) empathise with their client because with every word the client utters, 175 billion parameters are being analysed against every previous written or spoken instance? One that can take that information, assess, and summarise it in a few seconds and then present the response mimicking the style of whoever appears to have been most successful in similar situations historically?

We are not there yet and the CEO of OpenAI has indicated people are likely to be disappointed if they think version 4 will catapult us into an Ex Machina scenario. However, the level of excitement being created by glimpses of what is possible, is itself accelerating the progress, as vast sums of money and motivation pour into the technology.

Feeling nervous? ChatGPT currently has limited knowledge of the world after 2021 so perhaps you should ensure you start keeping up on current affairs before you engage in dialogue with anyone.

Share this post