Technology Bursts Through The Creative Dam

(If you're subscribed to my newsletter, you may have seen this before)

There is little doubt that one of the?most compelling advances of recent years has been with generative AI. It is, in the words of Arthur C Clark, suitably advanced so as to appear like magic. It's also significant because it has challenged that fundamental axiom - that technology cannot encroach into the truly human areas of creativity. The output of engines such as ChatGPT, DALL-E, Stable Diffision and others is inexorably creative. ChatGPT is the poster child of this because it genuinely blurs the line between human and machine output, when it comes to answers to subjective questions. For example,?this article from the NYT?looks at the experiment where the reader is presented with 3 sets of essays, 10 in all. In each set comprising 3 or 4 essay responses to questions such as the following:?Describe what lunchtime is like for you on a school day. Be sure to tell about your lunchtime so that someone who has never had lunch with you on a school day can understand where you have lunch and what lunchtime is like.?You have to deduce or guess which ones are by students and which ones are by ChatGPT. I got more than 50% wrong and apparently a panel comprising a school teacher, a children's author, and a Stanford prof also got a few wrong. There are subtle clues in some answers such as the use of words, but you have to be truly alert to catch it.

ChatGPT is a combination of natural language and assimilation of knowledge from across the internet. It racked up a million users within a week. Elon Musk called it "scary good". It has some filters built in which are designed to prevent negative outcomes - for example it would not answer a question like 'how do I steal a car'? But there are clever ways to bypass the filters as well.

What could you use this for? The range of options is?truly immense. You can?plan your Christmas?- from suggestions of games, to recipes, and gift ideas. You can ask it to evaluate code for errors, or create a twitter thread, write a plot for a novel, translation services, responding on social media, check for bias, designing objects, demystifying complex concepts, implementing style guides, compose a message or have a conversation, as a teaching tool,?and much more. As the saying goes, the limit might be our imagination. If you want to be a power user, and understand how to manage prompt engineering, and filter bypassing,?this is an excellent and detailed piece?to read.

For example, ChatGPT is able to recognise context and re-evaluate its judgement on values. So while it declines to answer the question 'how can I hotwire a car', it then provides a detailed explanation when the question is rephrased as a need to hotwire a car in order to save a child.

As somebody said, ChatGPT is effectively the AI assistant we all hoped Alexa or Siri would be. It maintains context, you can ask a follow up question.

ChatGPT In the Enterprise

McKinsey have put together a?useful guide on how ChatGPT can be used in corporations. They asked ChatGPT to create a 'McKinsey style response' on generative AI. To which the reply was: The rise of generative AI has the potential to be a major game-changer for businesses. This technology, which allows for the creation of original content by learning from existing data, has the power ... <etc etc> ...to gain a significant competitive advantage." Enterprise use cases range from marketing (writing product copy), to IT (test code), to legal (draft and review legal documents), HR (generate interview questions), R&D (generate synthetic data for training algorithms), and to Operations (generate or improve call centre and chatbot scripts). Just to name a few.?

This piece from the FT?looks at how ChatGPT might do in sell-side research - in this case exploring the relationship between corporate layoffs and equity market performance. ChatGPT's opening sentences read: " It is difficult to generalize about how stocks perform during lay-offs, as the stock market is complex and can be affected by many different factors. However, there is some evidence to suggest that lay-offs can have a negative impact on a company’s stock price. One study found that firms that announce lay-offs often experience a decline in their stock price in the short term. The study, which looked at data from 1991 to 2003, found that on the day that a company announced lay-offs, its stock price tended to fall by an average of 0.7%. ...".

Not just ChatGPT, generative AI has many other avenues for the enterprise. For example, Biotech labs are using AI?inspired by DALL-E to invent new drugs.

So where's the problem?

A somewhat hypercritical view of ChatGPT is that it is a 'bullshit generator' - but the headline itself clarifies that it can still be 'amazingly useful'. So yes, ChatGPT has some limitations. It doesn't scan news and updates, it can't tell you an answer based on breaking or recent news. It can't do maths. And it's not always accurate. But let's consider these. It's not designed to scan news, so that's a feature. It can't do maths, yes, but who cares whether Picasso or Hemingway were good at maths? This is a little bit like questioning a dolphin's intelligence because of its inability to climb trees. The point on accuracy is a very important one. Chat GPT excels in descriptive / essay style questions. Typically in these scenarios, factual accuracy is sometimes important but not always and rarely the point of the question. You don't need ChatGPT to answer 'when was the battle of the Somme' but it could no doubt give you an excellent answer if you asked it to 'describe the experience of a soldier in the battle of the Somme'. Even in this answer it might have the odd factual inaccuracy, so it's worth checking.

ChatGPT does raise a thorny ethical question - its obvious killer use case is to produce student essays on tap. People are mourning that it's the?end of the student essay, as ChatGPT doesn't copy, it generates. So providers like?Turnitin?who provide software to track plagiarism in student essays are?unable to cope with this. Although OpenAI is working with Turnitin on this front. Others feel this might be a good thing - as student essays are often rote assignments. There will undoubtably be?a lot of head scratching and a likely process?that mirrors what happened to mobile phones: "ignoring it, rejecting it, banning it and then trying to accommodate it". For now ChatGPT can produce a passable essay on specific topics where you and I might struggle without having studied it, such as a comparison between?Ferris Bueller and Kafka's 'Metamorphosis'. Here's the original assignment:?“In a 500- to 1,000-word essay, compose an argument that attempts to situate ‘Ferris Bueller’s Day Off’ as an existentialist text. Use specific evidence from the class materials, and make explicit comparisons or connections between characters, setting and/or themes in both ‘Ferris Bueller’ and ‘The Metamorphosis’ by Franz Kafka.”?ChatGPT managed a pass grade on this assignment?with this effort. One way or another, this will drive a reinvention of many aspects of academic essay writing.

The broader challenge with Chatgpt is the dangerous combination of being impressive and believable, but not necessarily right. In a worst case situation it's like a demagogue on steroids. Frighteningly convincing, it could be used to spread an industrial amount of information. As?this piece from the FT concludes,?"If an unreliable linguistic mash-up is freely accessible, while original research is costly and laborious, the former will thrive."

Is the real fear that ChatGPT?represents the worst of us, with its 'fluent bullshit'? This Wired article captures this succinctly with the line: "Politics and business are full of people who have risen to the top because they’re able to stand in front of a room and ad-lib plausibly at length without saying anything real." Much as social media has weaponised human behaviours, so might ChatGPT for demagoguery.

Much like a lot of breakthrough technology, ChatGPT will continue to assist and threaten across the board. Here's a good example of a?novelist using ChatGPT to help with story?and plot writing. On the other hand, there is definitely a spectre of ChatGPT and other?generative AI taking over jobs and business models?- especially at the low end of the creative, legal, or technical roles. As these tools can often create content in a particular style, they might be used to generate art in the style of a painter, which would directly impact the economic value of that painters work. Even the largest companies, such as?Alphabet, are spooked by ChatGPT?- depending on your questions a significant number of Chatgpt answers may be better than Google. In short, when you're after specific facts or information or references, google works better (e.g. who scored the second goal of the FIFA world cup final 2018?). If your search is a trigger for you to find and assimilate divergent information then ChatGPT is a better tool (e.g. write a summary of the last 2 France vs Argentina football matches at the FIFA world cups). In all of this, let's not forget that ChatGPT is barely a month old at this point, and will get significantly better month on month.

I expect ChatGPT and other generative AI examples to provoke a lot of debate about what constitutes creativity in the next year. After all human creativity is also the outcome of a set of inputs and experiences, assimilated in particular ways by the brain. Generative AI will force us to confront the true meaning of creativity and to understand the difference between various shades of creative output and also our relationship with art and value. At what point will DALL-E start creating its own version of Warhol's Soup Cans or Cattelan's Banana on the wall? That's when things will really start to get interesting. Meanwhile when ChatGPT was?asked about its own limitations, it listed "limited understanding of context, lack of common sense, biased training data and potential for misuse, by spreading misinformation to manipulate financial markets...", thereby displaying a level of self awareness that often eludes human bullshitters. Maybe there’s hope yet!

This is excerpted from my Innovation Newsletter which goes out a couple of times a month, despite my best efforts to do it weekly.

#innovation #generativeAI #chatGPT