Is Generative AI Overhyped or Underhyped?
Everyone has an opinion on Generative AI. But is the hype for real? We weigh in on whether the underhyping or overhyping are for real.
Seth Godin thinks generative AI is underhyped and will reshape our world.
Gartner says generative AI is about to enter the “trough of disillusionment.”
Who’s right when it comes to the generative AI hype??
They both are.
And, why does where generative AI is going matter for vector databases? As Andreessen Horowitz says, “The most important piece of generative AI architecture, from a systems standpoint, is the vector database.” How can they help?
Let’s explore why the state of generative AI is tough to pin down and what that state means to the development of vector databases.
Why Generative AI is Overhyped: The Science of Hype
Generative AI is overhyped because that’s how humans work: The behavioral economics principle of Planning Fallacy characterizes the extent to which humans are naturally overconfident. For example, in a 1994 study, a group of psychology students were asked to estimate how long it would take to finish their senior thesis. Their average estimate was 34 days; the average completion time was 55 days. Only 30% finished in the amount of time they predicted.
And the planning fallacy cuts both ways. When things are bad, we extrapolate our gloom and doom. For example, our long-term outlook becomes overly bleak when the stock markets decline. We over-hype some things and under-hype others.
Throughout time, this over-to-under-hype cycle is why Gartner’s famous curve is so compelling. Get into business as the hype is rising, assert your leadership, and you’ll be the winner, so the theory goes. For generative AI hype, we’re riding a hype cycle bigger than a wave during a winter storm at Nazaré. One measure is Google searches for generative AI, which went from a 3-5 on the hype Richter Scale to 100 in one year.
Using Generative AI in the Enterprise is Harder Than Most Think, But It’s Early
The game has just begun for generative AI, but many of us are fumbling the ball on our first possession. Most companies have rushed in to try AI, and most have found small wins, not the earth-shattering ROI promised. We’re starting to doubt ourselves. This is why we may be about to enter the trough of disillusionment.
But we must keep the hype in context. ChatGPT was released to the public one year ago, on November 30, 2022. In five days, it had attracted a million users and is generally regarded as the fastest-spreading technology innovation ever seen. For example, it took 62 years for cars to be adopted by 50 million people, and the telephone took three years to be in the homes of 50,000 people.
The hype of the iPhone as an internet device is a good analogy to ChatGPT. It took seven years for the iPhone to capture 20% of its current-day internet-surfing data volume. Most don’t remember the disillusionment of those days. Browning the internet was a terrible experience in 2007, and the first apps in the App Store were slow, buggy, and mostly unusable.
In other words, in just one year since the launch of ChatGPT, just like the iPhone, we’re historically really early in the game for generative AI. If the generative AI market was a Broadway play, we’re at the stage where the curtains are just starting to rise.
Is Generative AI Overhyped? Not Even Close
Bestselling author Seth Godin thinks AI is underhyped. He credits AI's scale, speed, and price as the reasons it’s not going anywhere anytime soon and likens it to the evolution of the cell phone.
According to Godin, at the beginning of the cell phone era, we used mobile phones for an average of 4 minutes daily. Today, we use them for 4 hours every day. Godin argues that GenAI will always be with us from here on out, just like the mobile phone and the internet. He argues that AI’s persistent presence in our lives is the key reason it is underhyped, and its disruptive power remains misunderstood early on.
Yet a recent survey I’ve conducted with my students suggests that 95% still use generative AI just a few times a day. Like data use on the iPhone, serious use is coming slowly. So, Seth’s four-minute mark still seems far away.
The Business State of the Union of Generative AI
As for business impact, the hype-or-overhype jury is still out, too. Marc Andreessen agrees with Godin that “Much as the iPhone revolutionized our daily interaction with technology—spawning products like Uber, DoorDash, and Airbnb—generative AI will change everyday life."
Consulting firm BCG reports that generative AI use cases deliver content and visualizations three times faster than traditional techniques and that GenAI is accelerating task automation at greater than 50% while producing engagement rates twice as high for personalized recruiting messages.
But the general business population isn’t seeing these kinds of results. Reuters reports that prominent AI platforms like ChatGPT and Midjourney have seen month-over-month declines in visitors. GenAI writing tool Jasper underwent a round of layoffs in July after experiencing declining user growth for four consecutive months. A Teradata survey found that 57% of executives at large enterprise companies believe interest in GenAI will fade, despite 89% of them saying they understand the merits and potential of GenAI.
So the jury is out on generative AI and furiously using ChatGPT to make up their minds.
Vector Databases: One Key Element of Generative AI Adoption
What role do vector databases play in the evolution and adoption of generative AI in the enterprise?
Andreessen Horowitz says that vector databases are the most essential element of generative AI architecture. Why do they make this claim?
AI models, especially machine learning and deep learning algorithms require vast quantities of data for training. As relational databases power spreadsheets that manipulate structured data, vector databases power the processing needed by generative AI: model training, similarity search, and real-time scoring.
In Vector Embeddings 101: The New Building Blocks for Generative AI, Nathan Crone explained three ways vector databases power generative AI:
Vector databases provide an efficient way to manage modern, unstructured data. Some of the most powerful generative AI use cases include the processing of unstructured data, such as social media posts like this one, videos, audio, and images. Vector databases use machine learning algorithms to identify similar features of these unstructured data types, group them, and assure quicker retrieval than a traditional relational database.
Vector databases provide AI models with historical data management for training and analysis. Vector databases excel at efficient data retrieval, particularly with time-based queries. This capability is crucial for AI algorithms to access relevant historical information quickly, enabling them to make informed recommendations.
Vector databases allow AI applications to be deployed for real-time applications that make predictions based on continuous data streams. Some vector databases are designed to ingest and process real-time data, ensuring that AI models always have access to the most up-to-date information. They enable AI models to access historical data and apply anomaly detection algorithms to identify unusual patterns or events.
When vector databases are integrated into AI development and deployment pipelines, data scientists and AI engineers can access time-series data stored in these databases when building and deploying AI models, facilitating the development process.
Just as the early iPhone required great internet coverage to make its apps usable, vector databases are the key to ushering in a new era of generative AI innovation. They simplify data management complexities, allowing data scientists and engineers to focus on refining AI algorithms rather than grappling with data infrastructure.
Healthy Skepticism Coupled With an Unwavering Belief
Generative AI is experiencing moments of skepticism coupled with unwavering belief in its transformative power. Vector databases are not just a technological advancement but a catalyst that will propel AI into our lives as persistently as mobile phones.
The symbiotic relationship between AI and vector databases, like the iPhone and the internet, will shape the future, driving innovation and transformation across industries and ushering in an era where AI is an integral and enduring part of our daily existence.
Vector Database Central is a reader-supported publication sponsored by KX. To receive new posts and support our work, consider becoming a free or paid subscriber.
Subscribe to Vector Database Central
Essays for senior leaders about the business of vector databases, vector programming, and vector analytics. {The, Weekly, Vector} newsletter.
Check out these Substacks we love: