Will ChatGPT change the world? Lessons from previous waves of automation software

Will ChatGPT change the world? To answer that we need to look at the adoption of previous generations of automation technology. How fast have they been adopted? And by whom?

Getting Paid

I've had trouble with late payers recently. More than usual. Sadly, after 17 years running my own businesses, I've come to expect a few invoices to be overdue here and there. But this was a lot of invoices, very overdue. It was stressful. For both me, and my clients.The people I work with in client organisations rarely have much control over what gets paid and when. Even if they are senior leadership. They can chivvy and chase, but they can't make it happen. It's embarrassing for them.By way of explanation or apology, clients have sometimes let me in on the internal machinations preventing me from getting paid. These stories are no longer shocking, but the interminable bureaucracy of large organisations* is always disappointing.The conversations bounce between time zones and departments. Finance, IT, operations, marketing, procurement, HR. Many people are involved. Ultimately, the problem is usually simple. Two digits are transposed in my bank details and somehow IT have to get involved to deal with a change request. Someone just missed an email, or was dealing with a backlog of work and holding up everyone else.There are no good excuses for late payment. But occasionally I have a little empathy. How much must these failures cost the business? How many hours of time? Across how many people? How much lost hair and how many sleepless nights?

Bug or feature?

You won't always see it as a supplier, but most companies now have some sort of process automation wrapped around their procurement. In fact, around most of their processes. The idea of these systems is to streamline things, make them more efficient, and avoid errors. Or is it?The evidence would suggest they're not very good at getting suppliers paid on time. Judging by the feedback from Twitter followers when I whine about late payments, my experience is pretty universal. And failures seem to create enormous inefficiency at the client end, costing them time and money.So what are these systems really for? And why do companies keep using them?One reason is to control spending. Leaders naturally want to know where their company's money is going and who is spending it. Another reason is fraud prevention - or at least ensuring that there is a good paper trail to prosecute those who do commit it. But couldn't they do these things and make things more efficient? Certainly, that would have been the promise of whoever sold them the system.

Humans failing the technology

The answer is that of course the system could be more efficient, as well as creating the appropriate controls and paper trail. But it's not (usually) the system that prevents that from happening. It's the people.Who benefits if the system is more streamlined? Headcount can go down. Fewer people are needed in finance and IT. The business might benefit but those people don't.Bosses would have more time on their hands. But they would manage fewer people. Have a smaller budget. They would lose prestige. They don't feel like they would benefit.This isn't to say that people are consciously sabotaging the system. They're not. They're doing their job. The same job they did yesterday, and the day before. Things that feel right, good and rewarding. They are making sure the processes are followed. Making sure no bad transactions get through. That's what they're there for, right?What they are rarely incentivised to do is fix the broken bits of the process and do themselves out of a job. And so, they don't.In fact, even when a new system comes in, they tend to keep doing what they did before. They bend the system to their old behaviours. Without enormous and often disruptive interventions in changing behaviours - and sometimes people - as well as technology, things largely stay the same, just with new software.

Is ChatGPT any different?

While all this late payment shenanigans was going on, my timeline was filling up with examples of the latest iteration of OpenAI's work, ChatGPT. A machine learning system trained on a sea of data to create new things based on simple text instructions. It can write a story, a press release, code or even a multiple-choice adventure game.At first glance it looks like some kind of doomsday device that will destroy employment in a variety of sectors. But analysed in the context of my experiences trying to get paid, I'm not so sure.Process automation technologies like the ones that should have seen me paid on time with the minimum of fuss are not new. And yet years - decades - after their introduction, human beings are still stopping them from delivering on their promise**. Out of self-interest, lack of interest, lack of incentive or support, we've stymied much of the promise of efficiency.Why should ChatGPT be any different? In many corporate contexts, I suspect it won't.

Beyond the corporation

Outside the walls of the corporation or other large bodies though, the situation looks very different. ChatGPT and its brethren are weapons of wicked efficiency for the lean - and the potentially unscrupulous.Twenty-something years ago I was working on the marketing efforts of a large US software firm. Let's just say they were involved in video, for fear of upsetting any old clients. The company's revenues always seemed slightly out of kilter with the small number of case studies we were ever able to offer. A few high profile sports leagues and a couple of broadcasters didn't seem like a significant enough customer base to justify the numbers they were doing. The reason, we all knew, was that the biggest chunk of revenue came from the adult industry. And no-one wanted to talk about that.Who was the earliest to latch on to the potential for streaming media? It was the adult content providers***. The same group were very early to the potential for the tablet. I heard an anecdote from someone in the adult industry that the day the iPad launched, the wholesalers (yes, pornography has wholesalers) were ready with all their content refactored to the appropriate screen sizes.Translate this behaviour to the here and now. Who will be the companies making the most out of ChatGPT?

High volume, tight personalisation, low quality

ChatGPT doesn't turn out amazing quality writing. Yes, it's better than a lot of first drafts I've seen from many writers. But it's not going to win a Pulitzer, or even get past any decent newspaper editor. If you have high standards for quality control, you're not going to be using it. Or at the very least, you will be doing a lot of editing before you publish.But there are a lot of places where quality is much less important than volume, and tailoring. Anywhere you want a lot of words about a particular niche, ChatGPT will be useful. As will equivalent platforms: while OpenAI might have content moderation, other platforms do not.Two industries spring to mind. The adult industry first of all, just like in the old days of streaming media. Sure enough, there is apparently a thriving community of people using large language models (the generic term for tech like this) to write niche erotic fiction. Whatever your particular peccadillo, you can now get an endless supply of tailored fiction to meet your needs. The web will be awash with it soon enough, as it will with its image, video, or interactive equivalent. Combine an AI-written script with deepfake tech and you have generative pornography.Then there is the search engine optimisation industry, and the web content industry more broadly. Tech-savvy people who are trying to maximise the return on investment of their time. Want to create a website that looks like the authority on any particular topic? ChatGPT could be your answer. Of course, Google's algorithm could be tweaked to spot AI-generated content (read enough and it has a noticeably idiosyncratic style). But that's just the latest round in an ongoing battle between those building websites and the businesses trying to help us navigate them.

Many niches

These aren't the only applications where ChatGPT will be successful. There are likely many more niches where cost and tailoring are more important than outright quality. But I think they are representative. And in every niche, this generative technology presents the same problem: navigation.We are already struggling to navigate the digital world. There's too much of it. Too much content. we struggle to choose the best use of our time, facing constant FOMO. What happens when there is a 10x, 100x, 10000000x increase in the volume of content out there? Without a meaningfully better way to filter and navigate, we will be lost, swamped by it.Technologies like ChatGPT will not just swamp us in content, they will drive us into ever smaller niches. Arguably the reason that so much investment goes into franchises like Marvel and Star Wars these days is that they are some of the few brands that can still attract a mass audience. With so much choice in front of us, it will be easier and easier to sink into a niche of one. That might sound appealing but it's not always healthy. We need content that connects us and creates a shared conversation. And I'm not sure a robot is going to deliver that any time soon.##*A journalist on Twitter asked for a good book about 'business' the other day. I pointed out that there's often very little difference between the internal workings of large organisations and the state organisations with which he is more familiar. This is more true than many in the 'dynamic, efficient' private sector would like to admit.** Note, it's not just humans. Some of the software is just crap.*** They weren't the only ones obviously, but they were operating at enormous scale very early. See also, online gambling.

What would you do if you could clone yourself? Meet my personal AI.

Imagine you could clone yourself. What could you achieve?What if there was a digital you, that knew what you knew. Knew everything you said or wrote. That was intimately familiar with your style. And that could respond to questions on your behalf. Or even help you to crystallise your own thoughts.I’ve been lucky to be one of the first people to get such a machine. Meet tom.personal.ai. Or tom for short.tom is a virtual me, primed with hundreds of thousands of words of my writing, from my books, my blog, and my social media. It’s learning more about me every day, feeding on my creative output from the last ten years. It’s beginning to absorb my podcasts, my presentations, and more.Now you can ask tom questions, just like my clients do.As an applied futurist I work with some of the largest companies in the world – brands like Pepsi, Mars, Google, Meta, Ford and BMW. They commission me to explore the future or to teach them to do it for themselves. But I only have so much time.With tom, I can open up my work to a huge new audience, not as static blogposts or ageing presentations, but as dynamic content. tom can be a virtual collaborator for a new range of clients, offering insight and inspiration on the topics that are relevant to them.What would you want to ask a futurist?This is just the beginning for personal AIs. It is a huge privilege to be part of the first wave of these robots hitting the internet, because it is a concept I have been talking about for over five years.I strongly believe that in the age of the metaverse we will need a machine that knows us. One that can help us to navigate the complexity of a world where the boundaries between physical and digital have fallen. And one that can help us to maximise our own cognitive capabilities – and our time.The company, Personal AI, that has built tom, is starting us down a road to where AI technologies can be really useful. Not for big companies but for us as individuals, releasing us to do more, create more, and have more fun.You can interact with tom now, over at metaverse.personal.ai. It’s early days so it’s still learning. Ask it some questions and see what it says. Share the answers!

A digital life after death

It is perhaps not surprising that during a pandemic, we find ourselves thinking more than usual about death and what lies beyond. So the news that Microsoft has filed to patent the process of building chat bots from dead people's social media histories, seems somewhat timely.Microsoft's patent covers chatbots built from anyone's digital history, not just the dead. But it is there that your mind immediately goes. Especially if you have been reading Neal Stephenson's latest novel Fall, as I have. Fall is about the creation of a virtual world into which human bodies can be scanned at the point of death. The newly created 'souls' retain some aspects of their personality, albeit not their full memories.

We can rebuild him

Microsoft's proposal is to delve through the digital archives of an individual and recreate their personality in digital form. The system would apparently draw on “images, voice data, social media posts and electronic messages” to build a profile. It might even use a ‘voice font’ assembled from recordings to make it sound like them, or recreate their image in 2 or 3D. Of course, with current levels of technology, we can't actually replicate human thought processes or capability. But call centre systems can already assemble original conversations from stores of data. A conversation with a chatbot such as Microsoft may seem fairly true to the original. It may even be able to say completely original things, if it can process news media through the lens of what it understands about a person's views.

Do not resuscitate

For me the main problem with this is about consent. Do you want a digital puppet based on your personality existing in the world after you are gone? Could you stop someone creating one if you wanted to? After all, many of us have sufficient digital footprints to support the creation of basic deepfakes today. It's not a massive leap to think someone could create a virtual clone of us today without our consent. The only reason I have a blue tick on Twitter is that someone - maliciously - created a digital identity pretending to be me, so some people are clearly motivated to do this. Imagine if that digital clone had been autonomous rather than human-controlled. Imagine if they could have spawned hundreds of virtual Toms, each time one was shut down.These are extremes. But we are already having to face issues of consent around digital resurrection. From holographic performances by Tupac or Elvis, to Kanye's holographic gift to Kim Kardashian: a speaking representation of her father with a message from beyond the grave. Who has the right to resurrect us?

The big conversations

These are questions to which we don't have answers today. Like so many technologies, this possibility creates questions for society about ethics, etiquette and law. And as is so often the case, we feel ill-equipped to address the range of questions at the speed required. Facebook may have abandoned its mantra of 'move fast and break things', but our approach to dealing with new technologies remains to break things first, then work out how to fix them.The prospect of a digital afterlife in one form or another is already moving from science fiction to reality. If you want to live forever, or want to ensure that your end is a true end, it may be time to update your will. 

A race between the four horsemen

In a recent post for for Locus Magazine, Cory Doctorow laid out his scepticism about general AI in a piece entitled 'Full Employment'. He argued that there is no sign that a general AI - one that can replicate human adaptability in tasks - is on the horizon. And that the work required to address climate change is so great that we are much more likely to see full employment than the AI-driven unemployment that many have predicted.I disagree with Doctorow's analysis of AI. Right now, I don't believe that we are close to a general AI. I am more open minded than Doctorow about the idea that current AI systems have the capability to 'evolve' into something more generally capable, but the gap remains large.My criticism is that I just don't think AI has to be very sophisticated in order to replace humans in the workplace. It's an argument that I have made many times on this blog, so I won't repeat it in too much detail here. Suffice to say that if you break any job down into its component tasks, today's machines are eminently capable of handling many of them. If you accept that machines take work - tasks, rather than jobs - then you can see that the remaining work can be redistributed among a smaller number of humans.Where I don't disagree with Doctorow is on the scale of the challenge presented by climate change. I have little doubt that large portions of humanity will be involved with the mitigation response. But the idea that this will offset any job losses due to automation brings me back to one of the most difficult parts of futurism: seeing not what, but when.

Four horsemen

Even before the pandemic, I was concerned about our prospects for the next 20-30 years. While it's not quite the apocalypse, there are four modern horsemen of disaster racing to cause us problems.

• Climate: In this period, directly or indirectly, climate change will start to affect the more moderate climates. Changes in weather patterns, disruption to agriculture, sea level rises. Until this point climate change has been something most people could ignore, should they so choose. This choice is going away in the next few decades.
• Technology: The prospect of technological disruption to employment and the economy is another major issue. Whether you want to generously call it AI, or prefer the perhaps more accurate 'machine learning and robotics', there is the potential for swathes of workers to be displaced by machines in the next three decades, from administrative, customer service, logistics and manual roles.
• Politics: We are in a rancorous period of global relations. Violence so far has been primarily inside borders rather than between them. But our international trading relationships are collapsing and our diplomatic ties being strained.  And domestic leaders in many countries seem to be incompetent, mad, corrupt, vicious, or some combination of all of these.
• Disease: The latest addition to the line-up is the global pandemic, spreading effortlessly through our international connections, strained as they are. It's unlikely to end quickly and we are likely to see more of its type.

The horsemen analogy falls down when it comes to timing. This isn't about which of these potential challenges will win a race to reach us. All four are here already. The question is the speed and scale at which their effects will be felt.

A race to the finish

Doctorow might be right. Our climate mitigation efforts might start well before we adopt robotics and ML technologies to a level that severely disrupts the labour market. Or he might not. The scale of job losses in the retail sector right now are pretty dramatic. We could attribute these to the pandemic, but really this is just the acceleration a trend towards automation and self-service that has been rolling for years. The pandemic may accelerate the adoption of automation technologies in the retail supply chain and logistics. It might also accelerate their adoption in other fields - administration, customer service, finance, law... Once people are out of the office, perhaps we will be less squeamish about replacing them with machines?Even if you ignore the technological effects, the pandemic has clearly had a terrible effect on our economy. Many are bracing themselves for  job losses in the coming months. During lockdown almost 150,000 people have been made redundant and over 9m have been furloughed. This doesn't even include the many self-employed who sit outside the support schemes or many not be counted as having lost their jobs, despite their income having collapsed. Full employment feels like a long way from here.This is especially true in the current turbulent political environment where it is hard to see coordinated efforts to restore global prosperity. Or for that matter, a coherent effort to address climate change. If we were to start that process now, I can see the creation of an enormous number of jobs that might redress the losses currently being experienced. But it feels more likely to me that these efforts won't start until the effects really start to bite. That is the nature of our politics right now: always focused on today not tomorrow.In the meantime, it is going to be a difficult few years, whichever of the horsemen is leading the race.

Robots are companions, not carers

I trust that the TV and radio producers I deal with are adept at gauging public interest in the stories on which they ask me to comment. So I’m sure there must be great interest in the concept of the care robot, a topic on which I was asked to comment three times yesterday, ending with a debate with The Guardian’s Michele Hanson on BBC Ulster.Michele and I were positioned slightly apart in our opinions, though perhaps not as far as it may have seemed to the listener. For while I think robots most definitely have a role to play in the care sector, I’m loathe to accept that they are in any way a suitable replacement for a human being.

Love technology, respect people

Any reader of this blog will know that I love technology. It has been my obsession from near-birth. But I also feel we are too appreciative of our own brilliance when compared to the spectacular complexity of our own bodies. We can’t yet understand nearly half of what we our bodies and minds can do, let alone replicate it. Rarely are those uniquely human characteristics more important than in a caring environment.For this reason we are a long way from having a robot that can ‘care’, however rapid the rate of technological progress. The revolution we require is not one of technology but of economics and social policy, properly valuing care work and creating a system to reward it appropriately. It’s hard to see how this will be achieved without radical political intervention in the economic system, something that might be decades away.What fits today’s system is an answer based on capital investment in technologies that can — if only in part — offset the lack of proper investment in humans in a care setting.This is where Michele and I differ. We agree that robots can’t care. But we disagree about whether robots can be useful companions.

Plug-in pets

I am not a pet person. Animals make me sneeze. Dogs scare me. And frankly it’s hard enough tidying up after myself and my kids, let alone adding an even less self-controlled creature into the mix.But I get it. I understand the appeal. I’ve seen the joy that animal companions bring to others. A joy that has been quantified by research. As the US Center for Disease Control, an organisation not prone to woo, puts it:“Pets can decrease your: Blood pressure, Cholesterol levels, Triglyceride levels, feelings of loneliness…”Now, what proportion of each of these benefits do you think is down to the innate capabilities of the animal? And what proportion is down to what happens in our heads through our interactions? The studies, though small so far, suggest that robot companions can offer the same benefits as living companions.Robots may not yet be even as smart as our pets. But they can be much better adapted to the needs of those they are designed to interact with. For a start, they can speak, tell stories, show films, control lights and heating, and clean floors rather than dirty them.Between these enhanced capabilities and our own propensity to anthropomorphise everything around us, it seems obvious to me that robot companions can be a useful supplement to human interaction.

Companionship is not care

This though, is the limit of their capabilities with today’s technology. Robots cannot replace humans in a care setting and nor should we accept that as a proposition. We have a major under-employment problem amongst the young, an ageing population and historical undervaluation of care work: one solution could hit the trifecta.If and when these problems are solved though, robots will remain useful and valid additional companions.