Let’s start our 2023 predictions off with some thoughts on artificial intelligence. With ChatGPT, Silicon Valley seems to have gotten a bit of its mojo back. After two decades spent simmering the magic of Apple, Google, Amazon and Facebook into a sticky lucre of corporate profit, here was the kind of technological marvel the industry seemed to have forgotten how to make – a magical tour de force that surprised, mystified, and delighted millions.
Even better, ChatGPT didn’t come from any of those corporate titans – not directly, anyway. Instead it came from a non-profit artificial intelligence research laboratory called OpenAI. Founded in 2015 with a mission of furthering “responsible AI,” OpenAI is backed by some of the most celebrated names in Valley technology – LinkedIn’s Reid Hoffman, PayPal’s Peter Theil, Tesla’s Elon Musk among them. Now this was more like it!
ChatGPT seemed to burst from nowhere – but of course, like Google or TikTok before it, its success leverages years of consumer behavioral data and decades of academic research in mathematics, artificial intelligence, and linguistic models. Over the past seven years, OpenAI has evolved its corporate structure to incorporate a for-profit model and more traditional venture investment schemes – with all their attendant complexities. Now owned in large part by the very investors who gave us tech’s last two decades of mixed blessings, it remains to be seen if OpenAI will remain true to its mission of ensuring “that artificial general intelligence benefits all of humanity.”
But let’s be real here: It takes capital to build at-scale AI applications – a lot of it. For all its tickling of the popular imagination, ChatGPT lacks a business model. And one of the most ironclad mandates of money is that money sown must become money reaped. Which takes us to the question driving my first predictions for the coming year: ChatGPT will drive several significant innovations in digital business models. The first will be for ChatGPT itself – it will start to license its technology to at-scale clients, initially to OEMs who will blend ChatGPT with their own offerings. The next two will come via ChatGPTs first big new clients. Google, which played an integral – if largely unsung – role in the technology behind ChatGPT – will launch a ChatGPT-like version of its core search offering. In enterprise markets, Microsoft, which invested a cool billion in the for-profit iteration of OpenAI – will launch a ChatGPT-inspired service aimed at its largest corporate clients.
Google: Conversational Search
“Oh sh*t, Google’s screwed.” That’s the consensus of scores of hot-takes on ChatGPT’s launch. “‘Code Red’ for Google’s Search Business,” declared the New York Times. The NY Post, naturally, took it further: “‘Scary’ AI ChatGPT could eliminate Google within 2 years.” And Casey Newton, one of my favorite tech reporters, said ChatGPT makes Google “feel positively prehistoric.”
Ouch. I tend to disagree with all those hot takes, but as the old Valley trope states, only the paranoid survive, and certainly ChatGPT’s success is a reason for the folks at Google to be looking over their shoulders. Or perhaps more fittingly, into the mirror, where they likely see a company that’s developed an unflattering middle-age paunch. Could it really be outrun by a smaller, more agile version of itself? Is that even possible anymore?
I’m quite sure the board and major investors in Alphabet, Google’s parent company, are not only asking these questions, they’re demanding answers. And those answers will most likely take the form of a new product from Google in 2023 that I’ll call “Conversational Search.” (If you’ve read this site for the past two decades, that term will certainly resonate!).
Here’s how I imagine it might work. Pairing the open APIs and source code of OpenAI (assuming the newly for-profit company will allow it), Google’s massive trove of voice data, and/or its own internal chat platform, Google will build a novel conversational interface to its flagship Google search application. Text-based search has always had what I call a “modal” problem: often the first answer to a query isn’t accurate. Many in the search field wish they could pop up a modal dialog after an unsuccessful search, asking “Did you mean…?” This would allow the engine to both refine results, and gather critical data that would allow it to better answer the query next time. But there’s a problem: More than 50 percent of users will abandon their search when they see a modal dialog box.
The ChatGPT model of conversational “prompt and response” solves for this problem, providing a fresh context for how humans like to gather information (in essence, by talking to each other). The company will probably dub its first efforts in conversation search as “experimental” – Gmail was famously in beta for five years – but this will be deadly serious project.
Plus, it’ll be fun. Imagine a mashup of Google’s high-fidelity search with the serendipity and human-like conversational tone of ChatGPT. Unlike the stilted voice prompts of Alexa, Google Home, or (shudder) Siri, Conversational Search would be like talking with endlessly wise, patient, and intelligent guru. Pulling such a feat off would take and extraordinary amount of work, CPU cycles, and scale, but…Google is capable of all that and more. Plus, Google is strongly motivated to figure out a business model for Conversational Search, and it’s the one company both most likely to pull it off, and with the most to lose if it doesn’t. Marketers have been crying out for brand-friendly innovations in digital advertising, and Conversational Search could be just the ticket (for more on that, I’ll link to a future post here, once I’ve written it).
Microsoft: Enterprise Explorer
Microsoft also has a consumer search business (Bing, anyone?) but the company makes its money in enterprise software, and it’s already in the business of selling AI solutions to big companies worldwide. What I’ll call “Enterprise Explorer” could be a hugely successful – and profitable – upsell to its top clients, who wouldn’t mind paying, say, another $10 million or so a year to have a useful, sexy, and energizing new application at their disposal.
So what would Enterprise Explorer (E2, to be corporate cute) be? Built, again, from a mashup of OpenAI technology and Microsoft’s Azure compute platform, E2 would address some of ChatGPT’s most annoying problems – its indifference to truth, for example, or the biases inherent to its Web-scale training corpus. The idea would be this: Train a specific ChatGPT instance on just the body of data owned or operated by a particular corporation. Most large companies have access to petabytes of internal data – everything from customer databases to internal messaging and document management platforms, all accreted over decades. Add in partner data – cleaned and secured through industry-standard methods like data safe havens – and you could hit a tipping point in terms of pattern recognition and results. E2 could spark a revolution in accessing, querying, and delivering enterprise- and industry-specific intelligence – finally paying off decades of empty promises about the power of digitization and “executive information systems.” Imagine every employee being able to – quite literally – ask the enterprise questions about itself. The mind…boggles. As with Google and Conversational Search, pulling off such a feat would require a staggering amount of innovation and work. And again, just as with Google, Microsoft is deeply motivated to do exactly that.
So to summarize, my first three predictions are this: One, that ChatGPT finds a business model, two, that Google launches an initially experimental Conversational Search interface, and three, that Microsoft announces or launches an Enterprise Explorer-like application for its major Azure clients.
This is the first in a series of posts exploring my 2023 predictions. Here’s a link to the second post, and the third. And here’s a link to the summary post. Thanks for reading!
2022: How I Did
Predictions 21: How I Did
2020: How I Did
2019: How I did
2018: How I Did
2017: How I Did
2016: How I Did
2015: How I Did
2014: How I Did
2013: How I Did
2012: How I Did
2011: How I Did
2010: How I Did
2009 How I Did
2008 How I Did
2007 How I Did
2006 How I Did
2005 How I Did
2004 How I Did