The breakthrough of AI is now

The Internet hit the mainstream in 1993 with the launch of the first graphical browser, Mosaic. Things really took off the following year with the launch of Netscape Navigator in October 1994. But despite the availability of easy-to-use Web browsers, it took time for the general public to get access to computers and Internet connections via modems, and to start e-mailing and surfing.

In April 1995, when we at Kairos Future began measuring Internet penetration among Swedes, no more than 1% had access to e-mail or the World Wide Web (www). By June of the following year, the figure had risen to 6%. It wasn't until more than four years later, in the fall of 1999, that the 50 percent mark was broken. Now we are witnessing the next technological revolution. But this time it's on steroids – and it's not about web browsing, it's about AI.

November 30, 2022 – AI's D-Day
On November 30th, ChatGPT was launched for a wider audience. Until then, language technology and big language models had been the domain of geeks and experts. Now the technology was available to everyone, and within five days, one million people had set up a test account with OpenAI.

News media, podcasts and, not least, social media have since been flooded with examples of what ChatGPT can do – and what it can't. Most people have been amazed at how capable AI has now actually become. Given one question or prompt ChatGPT can be made to sum up the essence of stoicism, explain how humans will be able to compete with AI in the future, write code based on plain text or Shakespearean-style poems – or even do near-perfect translations. Many have also tried ChatGPT's sister Dall-e, which can generate perfect digital images with a prompt, images so good they can win prizes in competitions. 

For most who have tested the abilities, it's clear that November 30, 2022 was a turning point. It was when history opened up and we all got a glimpse of the future. Advanced AI went from being something that mostly concerned techies and language engineers to becoming commonplace.

Simplicity beats all
At Kairos Future, we have followed, used and even to some extent contributed to the development of AI. In 2009, we started downloading blog posts and other information from the web and applying mathematics and statistics to find new and more efficient ways to analyse large amounts of news text, social media, patents, scientific articles and much more. And a couple of years ago, we lifted the technology leg into a separate company, Dcipher Analytics, which today has the most complete set of tools on the market for text analytics or language technology, what in technical language goes by the acronyms NLP, NLU and NLG – that is, natural language processing, understanding and generation.

So, for us and others who have been traipsing around the AI field for years, ChatGTP was, in a way, no big news. The big news was GPT3, the engine of Chat GPT, launched in 2020, a language model far superior to its predecessor GPT2. GPT3 quickly saw the launch of a range of services by companies such as copy.ai and Jasper.ai, which in one year raised $131 million in venture capital and launched a broad portfolio of solutions. With ChatGPT, the range of such solutions has exploded in less than two months.

From a technical point of view, ChatGPT was therefore not a major innovation. Rather, what was astonishing was the rapid spread of the chatbot in the general media, and the enormous interest it generated. But also, the fact that it had immediate implications for a wide range of companies and organisations. For example, some schools quickly banned home exams. The reason for this massive interest and impact is likely to be simplicity. Suddenly, talking and getting help from an AI was as easy as using Google's search box.

AI and knowledge work
However, the consequences of the technology known as Large Language Models trained on huge amounts of text – and on which GPT3 is based – lie ahead. And they can be expected to be as big for the academic and civil service sector – for information and knowledge work – as the steam engine, petrol engine and electric motor have been for manual work.

The father of management thinking, Peter Drucker, said long ago that the challenge of the 21st century would be to do to knowledge work what we did to manual work in the 20th century. Namely, to increase productivity fifty-fold. 

General language models have the potential to do just that. And it will happen fast, probably much faster than it took the steam engine and its successors to change the world.

Already today, GPT3-based services can:

• provide sensible responses to email, 

• analyse and summarise the day's email harvest, including attached pdfs, 

• read through tens of thousands of news articles and summarise the most important news, 

• create unique documents and reports, 

• write poetry and prose, 

• act as a sounding board for idea generation, 

• identify trends in data, 

• help us serve customers and citizens in a pleasant way, 

• generate social media campaigns, 

• create business ideas and business plans, 

• write code from plain text or 

• produce tutorials with screenshots and instructions. 

For better or worse.

Microsoft, the major funder of the research company OpenAI, will soon launch ChatGPT as part of the Office suite and the Edge web browser. OpenAI also plans to release a professional version without the limitations of the free version. It is also likely that its image counterpart, Dall-e, will be integrated into the same platform, allowing users to work with text and images in a single interface. And the next version of GPT, trained on hundreds of times more data, will be available soon.

Independent vendors, building solutions directly on top of GPT3 or using it as part of a broader solution, are now launching a torrent of services for every conceivable area. For Google Chrome, for example, plug-ins are already available to help with email writing.

Dcipher Analytics, as I mentioned earlier, has built its own infrastructure to quickly implement new AI models from different vendors and has a long list of custom solutions to problems that GPT3 doesn't cover. Decipher is now planning to launch custom versions of GPT3 as research bots directly in Teams and Slack. Their solution can also provide information about what the AI bases its answers on in the form of references, something ChatGPT does not provide. 

OpenAI is also far from the only player that is building or will build large language models. There is a race going on here between the big tech giants, with Google in particular seeing the risk of losing its dominant position in web search. 

Taken together, this means that in the very near future we are all likely to have access to lots of powerful general and specialized digital workers at the click of a button. Indeed, in some sense we will become, as we at Kairos Future have long predicted, managers with a staff of digital employees. It will also mean that we will need to learn new ways of working. For example, the ability to ask questions and formulate tasks with precision will become far more important. Exactly how much the digital workforce will cost is not yet clear but compared to the working time that can be saved, it is likely to be negligible in most cases.

Huge opportunities – and huge consequences
What we see materializing before our eyes right now, seemed to lay in the distant future in the beginning of November 2022. It was something we didn't need to speculate about now. We could do that later. But now "later" is here, and it is high time to stop speculating. We need to act.

The consequences for most professions, businesses and industries will clearly be far-reaching even in the short term. The productivity gains for individuals are already dramatic. Every day our social media channels are filled with testimonies of how many hours are saved just by using services such as auto-generating or auto-replying to emails or getting help to write articles and blog posts. Analysts using text analytics have long been able to complete research projects in a fraction of the time that previously took weeks and months, but this has required mastery of the technical platforms. For example, one of Dcipher Analytics' clients conducted a project where they reviewed hundreds of policy documents in PDF format in three Asian languages to produce a report on policy trends in Asia. It was completed in just over a week, a task that would have taken years without AI support. Now the thresholds are being lowered considerably and everyone is benefiting.

At the team, company and community level, there are also other and sometimes wider implications. One person in the Kairos Futures network noted that she is going to completely rethink her recruitment strategy. The reason was an experiment in which she asked her team of about ten communicators to write a short text on a topic and at the same time asked Chat GPT to do the same. Then the employees were asked to score the texts. At the top came two texts written by team members. Next came all the AI-generated ones. Last came the rest of the team members' texts.

Time for a unified approach
For some years now, it has been fashionable for companies and organizations to develop a digitalization strategy. But few, if any, have considered the opportunities presented by the leap in development that ChatGPT represents. What we are seeing now changes, if not everything, then very, very much.

Companies and individual organizations therefore need to understand today how services built on general AI models such as GPT3 can be used in different parts of the business, and how they can be used to improve efficiency and quality. They also need to understand how work within teams needs to be organized in order to fully exploit the potential of new tools, and they need to know what is required from the organization as a whole to successfully implement the changes required. Finally, they need to understand the ethical challenges and dilemmas that come in the wake of technology.

To wait and see will, in the light of history, very well be considered the sin of omission of the century.

Click on the image to enlarge​.

Solutions based on generic language models such as GPT3 have the potential to transform information and knowledge management along the entire chain from information gathering and analysis to communication and interaction with, for example, customers. They also change the way teams work together, opening new possibilities and requirements for the organization and management of structural capital, as well as interaction with external actors. 

What every manager, leader (and employee) needs to do now
An action plan for spring 2023 should consist of at least these four elements:
• Get your own experience. If you haven't already. Open a test account with one of the providers of GPT3-based services and see how they work. Also, spend a few hours getting up to speed on where the frontline is going and what's on the horizon.

• Identify opportunities and implications for your own organization. Where can existing tools be useful? How will different groups of employees be affected? What do your competitors seem to be doing and how might it affect you?

• Develop a strategy for how you can be proactive in the transition. For most companies, it is important to act quickly to avoid being left behind. This requires rethinking previous digitization strategies in light of the technological leap we are now experiencing. Don't make this a long shot, start now and think of it as an iterative process, with a strategic roadmap that will be updated quarterly or bi-annually over the coming period.

• Start experimenting. Start testing different solutions in your organization right away. Do this in parallel with a more holistic approach and deeper analysis.

To support the above work, Kairos Future is launching in the first quarter of 2023 a Future Project on the impact of AI on information and knowledge work, the domain that is now being significantly transformed by the arrival of large language models.

Want to know more about the project or get help with anything else related to the AI revolution – anything from lectures or strategy processes? Contact Mats Lindgren.

Read more about the project and download the prospectus here.

 

By Mats Lindgren