The 3 Month Old Bottle of Wine
AI is clearly having more than a moment. It's going through a generational change. But what generation is AI, how mature is it, where does it go next and when will it be ready to drink? In this article, I detail the generational stage of AI and how I expect it to progress from here.
First things first: Artificial Intelligence is a technology, not a product. It’s a first generation technology that is currently being used to enable single skill, single tasks specialist use cases in some products.
But first a quick refresher on how we got here:
Timeline
2015 OpenAI founded
2020 - ChatGPT3 first released
2022 - ChatGPT3.5 released
January to March 2023 - The world loses its mind for all things GPT
March 2023 GPT 4 released
100 Million users and > 1.8 Billion visitors per month
On March 21, 2023 Bill Gates published “The Age of AI has begun”
June 1 2023 - there’s no signs of growth slowing
What Generation is AI?
Generationally, AI is a 3 month old bottle of wine.
If AI is wine, it hasn’t matured enough to serve at a nice dinner and you’ll often want to spit it out. But if you drink enough, it will get you tipsy. All joking aside, it’s incredibly important to be mindful that AI in its now publicly available forms is a relatively immature new technology. Generative AI is a single flavor of what AI is and it’s currently creating the kind of buzz that used to be reserved for an Apple Keynote. Companies have been working on developing AI for over a decade and we’re just starting to see the first productizations of this technology.
Now many of you will read that and think “Adam is an AI bear” which is not accurate. I’m neither a bear nor a bull for AI; rather, think of me as being neutral. That may be hard to comprehend given how overly frothy many people are right now for AI. If I hear another “AI is the thing in the zeitgeist” comment I may have to give myself a time out. There are clear groups of AI bulls who want to infuse AI into every/all products despite suitability, applicability for use cases and good sense.
There are now many AI bears who are more than a little afraid of the existential threats and unintended consequences that unregulated AI could deliver to humankind. Note: many of the AI bears included the very companies that are making the most advanced versions of AI including Sam Altman. When Sam wears a suit, feel afraid of what he knows (and we don’t) and please be aware that he’s taking this seriously.
Now that we’ve established my neutrality, let me explain why I’m neutral. I am very excited by the potential of AI as a technology. Applied responsibly, it could free generations of people from mindless work that the last 30 years of software and applications has created. In a future state, it could deliver meaningful value as a co-pilot, complement or replacement for humans in many of the use cases that we’re frankly terrible at (think driving, operating heavy machinery, synthesizing data, undertaking critical health research etc). To be clear, I am not excited about shoving AI into every product. That’s not what AI is intended for and certainly not what product creators should be doing.
Conversely, I’m very aware of the current state of AI and recognize that what is commercially available today is a first generation reactive product. It doesn’t do anything unless we engage with it, tell it what to do and coach it to become more performant. It’s not artificial general intelligence (AGI) or anything even close to it. Part of my neutrality is due to this realization. The possibilities of AGI are stunning and in many ways mindblowing. Despite noises of some forms of AGI living in bots in Google’s labs, most experts agree that AGI is anywhere from 25-50 years away.
I am nervous about the arms race that is going on to add AI to many of the products that we use every day because it’s on trend as opposed to it’s useful, valuable or fit to purpose. History is littered with examples of technology being misapplied to products with truly awful consequences (think thalidomide, Tay, advertising on social platforms backing misinformation and countless more examples).
A core difference between many past misapplications of technology and current state examples of AI being misapplied is that AI has the potential to grow into something dramatically more powerful than any technology we’ve created with the exception of nuclear weapons. Personal feelings aside, I promised you an article on AI as a Generational Product so let’s get back to that.
How Mature is It?
AI is a first generation reactive product.
Said differently, most current productizations of AI are products that do nothing unless the user engages and performs a task with the product. Once the user engages, products infused with AI are able to automate specific tasks like writing, chat and spelling/grammar checks. Slack, Miro, Salesforce, Notion and more are all on the bandwagon and have baked their flavor of generative AI into their products.
AI is early in its product lifecycle.
The current generation of AI is able to handle single use cases and single tasks at a time. Capabilities are changing quickly and multi task single use cases will become normal soon for many AI infused applications. It’s super important to be mindful that AI is still in its introductory phase of AI’s product lifecycle. It’s taken OpenAI 7 years of massively intensive and expensive development to get to this stage and introduce a commercially viable version of their tech. I could speculate how long it will take to achieve meaningful scale growth and mature current state AI technologies but I honestly don’t know. What I do know is that shipping a generational technology is incredibly hard and massively capital and time intensive.
Where does it go next?
I am fighting the urge to write “teething pains” in all caps because it’s clear that there are massive headwinds swirling for current state AI technologies. The initial rush to bake AI into many products was inevitably going to lead to backlash as seems to happen whenever any product is wildly successful. But the headwinds facing AI are different. Expect governments to regulate AI and establish a regulatory framework for approval of AI in sensitive industries, mission critical products and labor markets where AI could displace millions of workers. Expect consumers to vote with their dollars and weigh in to decide just what version, flavor and uses of AI they’re willing to pay for.
Also expect that the intense interest in AI will lead to massive investment and capex in furthering development. Ask any winemaker and they’ll tell you that no matter how much money you throw at grapes, time is the only thing that makes wine mature. Time plus investment will drive AI to mature.
Current state AI is usually applied to closed fixed data sets that provide defined edges and corners for tasks, queries and workflows. As AI evolves to be able to dynamically add new closed data sets and, eventually add realtime dynamic open data sets, we’ll start to see the sheer applicability of AI really accelerate.
The insurance industry is often left out of conversations about AI. I’m mindful that products don’t come to market if there are no insurance products that enable the product creator to ship and sell in the market with appropriate coverage. To my knowledge, the insurance industry hasn’t created AI specific policies to crystallize the potential liabilities, consequences and costs that could arise from negative outcomes tied to AI. You may be thinking “Insurance? That’s the speed bump that you see?”. Fair enough but look no further than autonomous driving and why it’s not everywhere. If the insurance industry lacks data, lawsuits and settlements, laws and regulations and strong signals for outcomes, they simply don’t create policies and sit out until those items are in place.
The other thing that I think comes next is a deep analysis by every industry and company in the world to assess what AI means to their business. I’m seeing this play out already with an absolute flurry of policy makers, executives, thinkers and professionals all engaging in meaningful conversations to figure out what AI means to them.
I’m thrilled to see the amount of conversations going on as it shows me that a) AI is being taken seriously b) the consequences of inaction are clear c) AI is recognized as a generational product and d) there is a clock ticking with urgency to understand AI.
When will it be ready to drink?
Current state of AI requires customers/users with a) strong technographics b) high experience levels c) high tolerance of flaws and d) copious amounts of patience. Mass market consumers likely won’t tolerate the flaws and limitations of current state.
Work to Be Done
The work to be done is:
a) mature AI generationally so it can become proactive and eventually evolve to AGI
b) develop productizations that help AI mature generationally
c) sort out the governance model for AI and what generation is palatable and in what form (Prescriptive generation is fraught with danger and existential threats)
d) educate everyone so AI literacy is mainstream (think The AI Education Project) and average people are at a minimum able to understand AI x AGI x other flavors
Maturing AI comes from further a) development b) training and c) technological advancement. Current state AGI player Sanctuary.AI can currently enable a human to teleoperate a robot remotely for the purposes of training that robot to undertake tasks. This is precisely the kind of training needed to enable the Learn<Research<Develop<Test<Release<Iterate<Rinse<Repeat cycle for AI. This will enable it to learn edge cases, user preferences and problem solving techniques based upon how a human operator solves the problem/behaves. Eventually, the robot won’t need the human to train it and it’ll become more performant than the human. But that isn’t going to happen any time soon as it’s at least 3-5 horizons away (10-50 years) which is ok as co-piloting robots is a wonderful interim step to initiating robotics and automation into many industries.
Steps like launching plugins to enable ChatGPT to ingest data, information and knowledge from the Internet will facilitate faster training, larger data sets and a faster pace of maturation of their technology. “Users have been asking for plugins since we launched ChatGPT (and many developers are experimenting with similar ideas) because they unlock a vast range of possible use cases.” OpenAI. Plugins and integrations to existing tools, systems and platforms will enable interoperability, increase the user base and provide the basis for increased training at scale for OpenAI.
How do I expect it to progress?
Pent up Demand
Market traction of ChatGPT is to me less a signal of how mature, useful and performant GPT4 is and/or more a sign that there is such pent up demand in the market for new technology that advances our capabilities and what’s possible.
BigTech capex has been focused on a lot of crap/crapex (web3, metaverse, VR, social networks etc) for the last 20 years that frankly isn’t exciting or has captured our imagination. I remember watching Minority Report in theaters in 2002 and being awed by the idea of transparent floating interfaces that you could touch. BigTech and Hollywood have built this pent up demand and then shipped stuff that doesn’t even come close to delivering excitement. Apple - I don’t care if the next iPhone has a dozen cameras. It doesn’t matter.
Once the pent up demand starts to subside (and trust me, it will), I expect that we’ll have a moment where product creators recognize that the work to be done is to implement the technology in viable products and use cases or create new productizations of the technology.
Imitation is the sincerest form of
We’re starting to see the early signs of BigTech fighting to “own” AI and its uses, commercial model, distribution channels and inclusion in other products. They clearly see the commercial value of AI even at this early stage and I full expect that we’re going to see a Microsoft AI, Google AI, Apple AI, Facebook AI, IBM AI, Dicks Sporting Goods AI etc as everyone wants to take a swing at ownership of the stack.
The Inevitable Shake Out
Like every product category, there’s only room for 1 winner and up to 9 other products that can carve out profitable businesses. Translation: expect many to try and most to fail at trying to own their slice of the AI market.
Once this plays out and many of the AI attempts have failed, I expect we’re going to see a consolidation of AI technologies into the usual SDK < Tool < System < Platform < App Store < Community paradigms that have become normalized in the last 30 years for new technologies.
What Gets Me Excited
For those of you who have gotten to this point of the article and are still thinking “Adam is an AI Bear”, please allow me to reiterate that there are aspects of AI and the potential of AI that get me very very excited. I’m a proponent of building Generational Products and advancing those products to a point where they become invisible. AI holds the promise to help realize that vision of truly invisible products and a post-product state across many product categories. That makes me very very excited. The opportunity to free people from being operators of products while still getting the benefit from that product is badly needed. That said, I’m a realist that current state of AI is early innings and much more mature versions of AI are needed to achieve the advancements that many products need. I for one am excited by the possibilities and keen to see the probabilities of advancements improve with each new release of AI.
Open it and let it breathe
However the market progresses, it’s certainly an exciting time in the technology world. I love the energy and excitement around AI and I’m cheering for product creators to find long-term viable uses of this new technology. Despite the lightly snarky title of this article, I do believe that we’re going to see various forms of AI come to market in a state that is ready to drink relatively soon. I’m excited to see this new technology progress and will write in future Whitespace Volumes with updates on what I’m seeing. Until then, go have some fun with DALL-E 2 and ask it to create a photo of you as your dog (see header for context…).
