Join top executives in San Francisco July 11-12 and learn how business leaders are getting ahead of the generative AI revolution. Learn more
New products like ChatGPT have captivated the public, but what will be the applications that will actually make money? Will they provide sporadic business success stories lost in a sea of noise, or are we at the beginning of a veritable paradigm shift? What is needed to develop AI systems that are actually workable?
To chart the future of AI, we can draw valuable lessons from the previous incremental advancement in technology: the Big Data era.
2003–2020: The big data era
The rapid adoption and commercialization of the Internet in the late 1990s and early 2000s has made and lost fortunes, laid the foundations for corporate empires, and fueled exponential growth in web traffic. This traffic generated logs, which proved to be a hugely useful record of online actions. We quickly learned that logs help us understand why software breaks and what combination of behaviors lead to desired actions, such as buying a product.
As log files grew exponentially with the rise of the internet, most of us felt like we were onto something hugely valuable, and the hype machine was running at 11. But it remained to be seen if we could actually analyze that data and convert into sustainable files. value, especially when the data was spread across many different ecosystems.
Google’s big data success story is worth revisiting as a symbol of how data made it a trillion-dollar business that transformed the marketplace forever. Google’s search results have always been excellent and inspire confidence, but the company couldn’t have continued to scale search — or all of the additional products we rely on Google for today — until AdWords monetization enabled. Now we all expect to find exactly what we need in seconds, as well as perfect turn-by-turn directions, collaborative documents, and cloud-based storage.
Countless fortunes have been built on Google’s ability to turn data into attractive products and many other titans, from a rebooted IBM to the new giant of Snowflakehave built successful empires by helping organizations capture, manage and optimize data.
What was just confusing chatter at first turned into huge financial returns. It is precisely this path that AI must follow.
2017–2034: The AI era
Internet users have vast amounts of text written in natural language, such as English or Chinese, available as websites, PDFs, blogs, and more. Big data makes it easy to store and analyze this text, allowing researchers to develop software that can read all that text and teach itself how to write. Fast-forward to ChatGPT arriving in late 2022 and parents calling their kids asking if the machines have finally come to life.
It is a turning point in AI, in the history of technology and perhaps in the history of humanity.
Today’s AI hype levels are right where we were with big data. The key question the industry needs to answer is: How can AI deliver the sustainable business outcomes that are essential to continue this incremental change for good?
Workable AI: Let’s put AI to work
To find viable, valuable long-term applications, AI platforms must embrace three essential elements.
- The generative AI models themselves
- The interfaces and business applications that allow users to interact with the models, which can be a standalone product or a generative AI-assisted back-office process
- A system to ensure confidence in the models, including the ability to continuously and cost-effectively monitor a model’s performance and to teach the model so that it can improve its responses
Just as Google unified these elements to create workable big data, the AI success stories must do the same to create what I call workable AI.
Let’s take a look at each of these elements and where we are now:
Generative AI models
Generative AI is unique in its wildness, bringing challenges of unexpected behavior and requiring continuous learning to improve. We can’t fix bugs like we would with traditional procedural software. These models are software built by other software, composed of hundreds of billions of equations that interact in ways we can’t understand. We just don’t know what weights should be set to what values between what neurons to prevent a chatbot from telling a journalist to divorce his wife.
The only way these models can improve is through feedback and more opportunities to learn what good behavior looks like. Constant vigilance regarding data quality and algorithm performance is essential to avoid devastating hallucinations that can alienate potential clients from using models in high-stakes environments where real dollars are spent.
Build trust
Governance, transparency, and explainability, enforced by real regulations, are key to giving businesses the confidence that they can understand what AI is doing when missteps inevitably occur, so they can mitigate the damage and work to improve AI. There is much to be applauded for the first steps taken by market leaders to create well-thought-out guardrails with real teeth, and I urge the rapid adoption of smart regulation.
In addition, I would demand that all media (text, audio, image, video) generated by AI be clearly labeled as “Made with AI” when used in a commercial or political context. Just like with nutrition labels or movie reviews, consumers deserve to know what they’re getting into – and I think many will be pleasantly surprised by the quality of AI-generated products.
killer apps
Hundreds of companies have sprung up in a matter of months, offering applications of generative AI, from creating marketing materials to creating new music to creating new drugs. ChatGPT’s simple prompt could potentially surpass the search engine of the Big Data era, but many more applications could be just as powerful and profitable across industries and applications. We are already seeing huge improvements in encryption efficiency with ChatGPT. What’s next? Experimenting to find AI applications that deliver incremental change in user experience and business performance is essential to creating workable AI.
The companies that will build their fortunes on this new class of technologies will break through these innovation barriers. They solve the challenge of building continuous and cost-effective confidence in the AI while developing great apps combined with solid monetization based on powerful underlying models.
Big data went through the same cycle of noise and nonsense. Likewise, it will likely take a few generations and missteps, but by focusing on the principles of Workable AI, this new discipline will quickly evolve to create a platform for incremental change that is as transformative as experts expect.
Florian Douetteau is CEO of Dataiku.
Data decision makers
Welcome to the VentureBeat community!
DataDecisionMakers is where experts, including the technical people who do data work, can share data-related insights and innovation.
To read about advanced ideas and up-to-date information, best practices and the future of data and data technology, join DataDecisionMakers.
You might even consider contributing an article yourself!
Read more from DataDecisionMakers