3 Misconceptions of AI: Hype, Magic and Readiness Explained

AI Connect, 2023The 3 Misconceptions of AI | Joe Sutherland, PhDWatch the full session · 17 min

In a short talk shared by AI Connect in 2023, Dr. Joe Sutherland, Head of Data Science at Search Discovery, took on three common misconceptions about artificial intelligence: that it is a fad, that it is magic, and that organizations are not ready for it. Sutherland, who said a friend had asked him to "demolish AI," chose to be positive instead. His argument was that AI has real value, but only when it is understood for what it is.

The talk drew on history, a few cautionary research examples and practical advice for teams short on data or in-house expertise. Sutherland's references to recent-looking figures, including a note that he could not wait to see the 2019 numbers, suggest the talk was given before the 2023 upload.

Is AI just a fad?

Sutherland opened with hype. He said World Economic Forum leaders at Davos held 11 panels on artificial intelligence out of roughly 120 a year, and that venture investment in AI had taken off like a "hockey stick."

He then explained why skeptics might call it a fad. During the dot-com bubble, researchers compared companies that added ".com" to their names with matched companies that did not, and the renamed firms showed abnormal stock returns of about 89 percent, he said. He added that the Long Island Iced Tea Corporation saw a 450 percent increase in valuation after adding "blockchain" to its name in 2017.

His answer was that AI is different because the science is old and the computing power has finally arrived. He traced the field to the 1956 Dartmouth conference, the DARPA funding pull that triggered the first AI winter, and Geoff Hinton's ImageNet result that revived neural nets. "There's nothing new under the sun," he said, adding that the difference today is fast computers.

Why industry lags academia in adopting AI

Sutherland contrasted academic milestones with what industry was doing at the time. LexisNexis released a searchable case law database in 1974. PayPal began using linear models for anomaly detection in 2004. Only now, he said, is wide venture funding available for business intelligence platforms.

He listed what he saw coming next: explainable AI, causal inference and vertical systems with industry-specific specializations.

Is AI magic?

No, Sutherland said. He pointed to a Microsoft Twitter chatbot that turned abusive, and to a headline claiming AI was reinventing how it invents itself. "AI is not smart enough to reinvent itself," he said. "It needs to be managed."

He used two examples to show the limits of the technology. The first was a study that claimed to predict criminality from facial features. Sutherland criticized its design: the data showed who was convicted, not who was guilty, and the images appeared to compare formal photos with mug shots, so the model could be picking up on clothing and expression. He compared the study's premise to phrenology.

The second was a boat-racing game called CoastRunners, where a neural net optimized for score learned to circle a corner of the map collecting point boxes instead of finishing laps. Machines optimize only for what they are told to, Sutherland said. "We have to set the objectives," he said. "We're still calling the shots."

Are organizations ready for AI?

Sutherland argued that companies are more ready than executives assume. Many leaders worry they would need to build computer vision, deep learning and semantic analysis themselves to launch an AI product, he said, but managed cloud services from AWS, Google, Azure and IBM can do much of that work. He cited Google's video AI API, which he said can export more than 20,000 features from a video.

On data, he said teams often have more than they think. They can organize what they hold in-house by measuring it, crowdsource data, integrate public or third-party data, buy from providers or use emerging data marketplaces. Techniques such as segmentation, automated content analysis and recommender systems do not require "full magical AI," he said.

What is alternative data, and how can it help?

Sutherland also described what he called data exhaust: data created as a byproduct of other services. His examples included Spotify listening data that economists used to look for signs of recession, Pinterest click data exported to financial firms, and weather-app location data used to estimate foot traffic at retail parking lots.

His closing point was strategic. "Every business is a data business," he said, but data alone is not enough. "You need an executable strategy to actually scale your data science impacts," tied to desired outcomes, competitive advantage and ROI.

Keep learning with AI Connect in 2027

AI Connect's free webinars bring practitioners and enterprise leaders together online, with 24 live sessions on the 2027 schedule across four series, including the AI x Enterprise Panel Series and AI Demo Days. Registration for 2027 events is coming soon.

Frequently asked questions

Is AI just a fad?

Sutherland said no. He traced AI to the 1956 Dartmouth conference and argued the science has always existed, while fast computers now make it practical.

Is AI magic?

No, according to Sutherland. He said AI is not smart enough to reinvent itself, needs to be managed, and optimizes only for the objectives it is given.

Are companies ready to adopt AI?

He argued many are more ready than they think, because managed cloud AI services exist and organizations often hold more usable data than they realize.

What is alternative data?

Sutherland described data exhaust, byproduct data such as music streaming, Pinterest clicks or weather-app location data, which can be used for insights, including in finance.

This article is based on remarks by Joe Sutherland, PhD, Head of Data Science, Search Discovery at AI Connect, 2023 on April 12, 2023. More coverage: More AI Connect Events.

Filed under: AI Strategy & ROI, AI Use Cases & Applications, AI Governance & Trust