Wife: Have you heard of Chat CGT?
Me: Chat CGT? That sounds like a telephone based tax adviser service (little chuckle to myself)
Wife: (not even a smile)
Me: Do you mean ChatGPT?
Wife: Yes, that’s what I said.
Me: Of course, I been using it a little bit over last few months
Wife: Oh, I see. Is that who is writing your articles?
Me: No. I write them, thank you very much! But I have used it to help with some of the research.
Wife: (doesn’t believe me) Anywaaayyy… apparently you can type in any request into ChatGPT and it will write out the answer. Do you think it could do my job instead of me?
Me: It’s not at that stage yet, but will definitely disrupt and change how we work
Wife: You seem to know a bit about it. What does GPT stand for?
Me: Er…Er… (tries to secretly search for the answer)
Wife: I can see you looking it up!
Me: It’s a fair cop. Perhaps I need to dig a little deeper on AI to understand the subject
Wife: Great, but don’t use this conversation in your article
Me: Of course not 😉
Taking on the task of writing about AI is a fairly daunting one. It is such a complex and fast moving subject, so fast in fact that by the time you read this article there will be probably another news story about AI dropping in to your feed.
If you feel anxious about the future I would steer clear from looking up ‘AI predictions’ in Google.. some of them read like the plot of the Will Smith film “I, Robot” (spoiler alert: the robots take over the city!)
Of course, human beings have been through significant technological transformations before. From the agricultural revolution to the industrial revolution, or inventions such as the motorcar, air travel, electricity, the internet, or the telephone which have changed our lives beyond recognition. Is AI just one more new thing that freaks everyone out for a while, and then we figure out how to live with it? Or is this time different? Is AI going to substantially change our lives? And if so, will it be for the better?
Artificial Intelligence (AI) models like ChatGPT gives users access to a Large Language Model (LLM) and are effectively large prediction machines. The LLM takes in a vast amount of data from books, articles, websites, and other text-based sources, and is then trained so it can understand human language, context and generate responses. The LLM then responds to our prompts by predicting which words a human is going to want to read next based on the context of the prompt. So, whatever you ask it to do — write a song, write a CV, interpret data, translate language or generate some software code, it will analyse billions of pieces of data, then the LLM will produce what it thinks you have asked it to do. The output is rarely correct the first time and will usually require more specific and varied prompts.
I found this a strange concept to get my head round at first, because the LLM learns from data rather than specific instructions, the LLM is just predicting the next word in the sentence with the highest probability. It’s not actually ‘thinking’ in a human sense.
Transformers: Chatbots in disguise
The term “artificial intelligence” was first used in 1956. Back then, scientists were intrigued by the idea of creating machines that could mimic human decision-making. Ever since, the burning question has been whether a computer would be as capable as a human. Up until very recently this still seemed a distant prospect, but the introduction of ‘transformers’ in 2017 was a game-changer. Transformers handle sequences of data (like sentences) effectively, which allow the LLM’s to understanding context more deeply.
Things are now moving quickly. The current best estimates of the rate of improvement in LLM’s show capabilities doubling every 5 to 14 months. But this growth is not linear. AI doesn’t grow in a steady way, but tends to make huge leaps forward. Today’s AI systems are very good at one-off small tasks but are still fairly narrow in their capabilities. The next group of AI systems will be bigger and more powerful, with an entirely new set of capabilities. They will suddenly appear in our lives, and we will need to understand the implications and opportunities, or as my children say to me “deal with it”.
This is what makes identifying the winners and losers in AI such a difficult task. First movers don’t always win. Google was launched many years after the original search engines, but still went on to dominate internet search. This uncertainty, however, has not prevented a boat load of money being thrown at anything AI related. Venture capitalists have invested more than $40 billion into AI companies this year alone. To put that in context, GPT-4 cost much less than $1 billion to develop.
In the short term I would expect this trend to continue, given the nature of this structural theme, but after substantial gains last year in a number of AI names, they could face setbacks if profits fail to meet high expectations. As we saw in the Dotcom bubble, over the longer term investors will need to be more selective about which AI sub themes to invest.
Meet the new boss, same as the old boss 🤖
One of the big concerns about AI is job losses. The CEO of IBM said that 30% of back office roles, nearly 8,000 jobs, could be replaced by AI. This kind of announcement creates a lot of anxiety about AI taking jobs. If AI programs become as capable as humans will we all be replaced by a computer program? This is a familiar story. A new technology comes along that can do something that previously had only been done by humans, and it’s cheaper for employers to buy the machines than employ someone. The term for this is in economics is ‘Creative destruction’ , when one kind of industry disappears, but is replaced by other kinds of industries. There have been many examples of this down the years, such as when the automobile industry grew rapidly and eventually made horse-drawn transportation obsolete. Blacksmiths, stable boys and carriage drivers lost their jobs, however, the birth of the car industry also gave rise to new industries such as constructing roads, motels, car maintenance. Additionally, it also gave birth to the leisure, travel and fast food industries, as people had more time on their hands and the ability to travel faster.
AI will undoubtably impact many industries, some more than others, and new jobs have started to emerge. Chief AI Officer? Yep, it’s a thing. CAIO’s oversee the deployment of AI within an organisation. ‘Prompt engineer’? Nope, me either. This is someone who is essentially an expert in helping human users interact with the AI system.
Another trend is companies assigning tasks to ‘AI agents’. Agents are AI systems given the ability to plan and use tools, allowing them to act autonomously, they can work effectively as “AI contract workers” who are delegated jobs. Organizations of the future may have many more virtual employees and less human employees.
Regulators, mount up
When a technology like AI comes along, which has such wide ranging potential implications, governments need to actively consider regulation. These regulations need to still encourage innovation while also addressing risks related to misinformation, job displacement, bias, and privacy. A recent example of this is the UK competition regulator beginning to probe the alliances between Big Tech companies and AI start-ups, and also draft legislation that would likely put limits on the production of LLM’s. It is imperative that guardrails are in place in the development of AI, but given the speed of development regulators will invariably be one step behind.
Despite all of the concerns, there are also many positives. We are already seeing improvements in productivity in many areas such as robotics, finance, customer service, cyber security and healthcare.
Healthcare is an especially exciting area. AI can revolutionise the industry from disease diagnosis, drug discovery, medical imaging, predictive analytics to personalized treatment plans. This has already started to happen. A recent example is OpenAI’s GPT-4 model almost matching specialist doctors in both analysing eye conditions and suggesting treatments.
This is not to say AI will replace doctors, as people will still want to speak to a human, but it could allow doctors to be more efficient, improve patient outcomes and reduce pressure on overworked healthcare workers. Who wouldn’t want this?
It is hugely difficult to predict just how and far AI will change our lives, but I can predict that it will continue to surprise us. Things that we thought a computer could never be able to do, will all of a sudden seem normal. As an investor I remain alert to the opportunities that will present themselves. As a father I am telling my children to learn about AI. And as a human I am trying not to freak out too much.
Me: Are you still awake?
Wife: Just about, but just in case I drifted off in the middle, can you summarise?
Me: Sure (unbelievable!) what part?
Wife: Will my job be taken by a robot?
Me: Probably not, but AI will eventually be embedded in the operations of most firms, so it’s worth playing around with it now.
Wife: Are you investing in AI?
Me: We have concerns that AI has become a very crowded area of the market, and the valuations may have gotten ahead of themselves, but over the longer term this will be a big structural theme to invest in. Given the eventual disparity between winners and losers we think an active manager can make a big difference in this area rather than buying a passive index.
Wife: 🙄(rolling her eyes) Is that a ‘yes’?
Me: Yes.
Me: (looks on google for a few minutes) By the way, GPT stands for “Generative Pre-trained Transformer”.
Wife: (Has already left the room)
If you are interested in hearing from actual experts on the subject, then the below blog and podcast are good starting points.
https://www.oneusefulthing.org/
https://freakonomics.com/podcast-tag/how-to-think-about-a-i/
Ft.com is also very helpful
