The AI Strategist - 04.08.23.
6 reasons why Generative AI is a different kettle of fish to the AI that came before it
Cold Hard Cash - Case(text) in Point
Casetext, a legal startup with an AI-powered assistant for law professionals, has been acquired by Thomson Reuters in an all-cash deal.
Founded in 2013, the business model journey for Casetext has been communities => exploring AI to automate workflows => using Generative AI for research, document review and contract analysis.
Six months ago there was a lot of scepticism around the disruption/value creation potential of building applications on top of foundation models. Casetext gained early access to GPT-4 (released March 14th 2023) and this first-mover advantage has paid off. Revenue in 2021 was $4.9m. This has risen to $14.4m. So there is a pretty exciting narrative here … the Open AI partnership led to product development which led to explosive customer growth which was a major factor in the acquisition. Optimists 1 - 0 Sceptics.
Narrow-moated-rated Thomson are clearly also doing this for strategic reasons.
Why this matters - it is totally logical that businesses such as Casetext that have the data, customers and ML infrastructure already in place are best positioned to capitalise on Generative AI and disrupt at pace. It also clear that moving quickly has created a ton of value. Pace > overthinking defensive moats and moving slowly. Business leaders, perhaps already frustrated with the ROI of traditional AI, need to quickly get to grips with what this could mean for their existing business model.
2. Leader Optimism
Based on a BCG Survey, respondents are optimistic about how AI (particularly Generative AI) will affect their work, feeling that it will save them time and promote innovation in their roles.
36% of Generative AI non-users are optimistic about it vs 62% for regular users. I am reminded about Bill Gates explaining The Internet to David Letterman.
Equally interesting is how optimism is split by job role: frontline employees (42%), managers (54%), leaders (62%).
Why this matters - the sentiment feels different compared to the 2019-2022 period I was working in ‘traditional AI.’ Back then, it felt more like a bottom-up movement where technical people were attempting to convince leaders of the potential and value. If we see more of a top-down movement with Generative AI this will have a dramatic impact on budgets, mandates and the pace of innovation.
3. Employee Productivity
At CHAPTR we are focused on the blend of human and machine intelligence - I have blogged about it before. In this fascinating report about the impact of ChatGPT on business writing tasks, productivity was substantially raised. The technology reduced the time required by 40%. Furthermore, the human review of AI-assisted outputs was rated 18% better than human-only outputs.
Why this matters - data is coming through showing us how ingenious humans are at experimenting with Generative AI to make them more productive inside and outside of the workplace. This trend will continue. Innovative and ambitious juniors will spot the opportunities to unblock and rethink workflows. They will also be knocking on the open doors of their increasingly optimistic leaders. This will accelerate adoption. It will feel more like a movement of time-urgent, frustrated & curious employees rather than the wading-through-treacle attempted IT implementations of 2015-2022.
4. It’s More Fun
In this enlightening and entertaining piece Professor Ethan Mollick explains how Generative AI is more fun. It is sub-optimal if many work-related tasks are boring but important. There is a very real business case here. People at work report being bored 10 hours per week. Business leaders have a responsibility to take this pain away and Generative AI will solve many problems here. I smiled reading about how much fun a University Professor in Philadelphia is having with Midjourney.
Why this matters - I use Midjourney a lot. For this blog, for work presentations etc. It saves me time and is more fun. I love the challenge of expressing my creativity and pushing the machine. I like the feeling of getting better at it (see autonomy, mastery, purpose). It is a million times more fulfilling than what came before. I don’t recall a time in the pre-Generative AI era that a client described any of our solutions as fun. We were building AI technology to solve a problem, or drop into a workflow, but it was nigh impossible to build a solution that an end user could co-create with in this way.
5. AI has Finally Become Transformative
In May I wrote about the 90% of organisations that have not had ROI with AI. I asked the question …
Will these organisations be brave enough to invest in Total Enterprise Reinvention to realise returns from Generative AI when they have had their fingers burnt with AI over the last 5-10 years?
Well Martin Casado from Andreessen Horowitz thinks it is worth it. He makes the argument that:
‘AI Has Finally Become Transformative - After a decade’s worth of innovation, new models can change the world the way the internet did.’
Here is an incredibly condensed version of his article. Earlier generation AI startups hire too many people too quickly, therefore building a human-in-the-loop solution around their own people, therefore they become low margin and hard to scale.
Why this matters - many of the challenges of implementing AI that came before (i.e. Human-in-the-loop, AI Safety, ML-Ops, existing infrastructure, change management) remain with Generative AI. The step-change, however, is that we are seeing obvious early evidence that the time, cost and performance of Generative AI use cases in established markets are ‘orders-of-magnitude’ improvements over what has gone before. Put another way, most organisations on the planet will have to find a way to make Generative AI work for them (whether that’s for internal or external innovation).
6. Throw Previous Use-Case Thinking in the Bin
I absolutely love the Generative AI use-case taxonomy breakdown suggested here. Imagine six layers of an onion. Layers 1-4 is a bit of an innovation playground right now. Layers 5&6 - serious players, massive disruption.
Advisory - here are my company financials, give me some tax guidance.
Assistive - book my travel based on a calendar invite.
Co-operative - human and machine in partnership to achieve something together.
Augmentative - help me achieve something at an expert level.
Digitally Autonomous - let the AI execute tasks and make decisions in a simulated world.
Physically Autonomous - autonomous robots.
Why this matters - a) it is nothing like the use case work I did before. b) it is so much more than just ChatGPT!
In a previous AI life I interviewed a procurement manager in e-commerce fashion.
Question - what would you do if our algorithm recommended that you should buy 500 SKUs when you were planning to buy 1000 SKUs?
Answer - I would buy 950.
We never proceeded with that prediction use case. Perhaps we should have. In 2-3 years, we will be using a framework like the above to sit down with the e-commerce manager with examples of proven co-operative use cases (3) to inspire them. Then design a Generative AI solution tailored to their culture and specific challenges.
In the Generative AI era we will be designing AI-friendly processes around human needs and behaviours, not brute-forcing unexplainable predictions into their existing workflows.