The Reality of AI: Insights from a Senior Analytics Executive

The Current State of AI Language Models

Patrick Vanguard, Senior Vice President for Analytics and AI at Sears, offers a sobering perspective on the current capabilities of AI language models like ChatGPT. Despite the impressive nature of their outputs, Vanguard argues that these models fall short in several critical areas:

Impressive Output, Flawed Content

Vanguard acknowledges that AI language models produce grammatically accurate, well-formed text with sophisticated vocabulary. They can generate content tailored to specific audiences, from kindergarteners to adults. However, he emphasizes a crucial point:

"If you actually read what it says, you'll discover that it's all wrong. The facts are wrong. It doesn't understand causation, it doesn't understand logical reasoning, it cannot do arithmetic, and it doesn't deal well with times."

This disconnect between form and substance highlights a significant limitation of current AI technology.

Comparison to Human and Animal Intelligence

Vanguard goes further, stating that language models are not even on the level of a human child. He references a clever wordplay in the AI community:

"Many people are talking about 'God-like AI,' and some of the great minds of AI have just turned it around and said we don't even have 'dog-like' intelligence."

He argues that even a dog possesses greater factual ability and intelligence than ChatGPT at present.

Practical Applications and Limitations

Despite these limitations, Vanguard sees potential for AI language models in certain contexts:

First Draft Assistance

AI can be useful for generating first drafts of various content types:

• Essays
• Marketing messages
• Emails
• Source code

However, Vanguard strongly emphasizes the need for human editing and oversight:

"The emphasis is on first draft. You will have to edit that. If you attempt to just compile and run the source code that comes out of ChatGPT, first of all, it will most likely not do what you want."

Productivity Gains

Vanguard estimates that using AI for first drafts can lead to a 20-30% productivity gain in terms of time savings. However, he cautions:

"You cannot automate anybody's job completely."

Current Limitations

Vanguard points out that AI is not yet capable of performing seemingly simple tasks with consistent accuracy:

"If what you want really is a digital assistant, somebody where you can speak to it and say, 'Okay, book me an appointment with you on next Wednesday at 3 PM,' and have it go away and correctly create a calendar invitation, send it to both of us, and have it be right... we're not there yet."

The AI Revolution: Hype vs. Reality

Vanguard offers a nuanced view on the claims of an "AI Revolution" made by prominent figures like Elon Musk and Steve Wozniak:

A Research Accomplishment

"It is a revolution, and it is scary. But we have to divide up what these things are. It's a revolution because we, the AI community, have managed to do something that we've been wanting to do for a long time and haven't been able to. It's worth a celebration, but it is a research accomplishment."

He emphasizes that current AI technology is not ready for widespread deployment in everyday work scenarios.

The Real Threat

Vanguard argues that the true danger of AI lies not in science fiction scenarios, but in its potential for misuse:

"It's scary not because it's the Terminator. It's very far away from The Terminator. But its primary use case will be propaganda used by authoritarian regimes to prop up their misinformation and disinformation. That's the number one use case in my mind."

Economic Implications

Vanguard points out that the economic benefits of AI are currently concentrated among a small number of tech giants:

"There are very few owners of this. There's about four or five companies that can afford and have, in fact, afforded to train models like this. Those people are going to make a great deal of money."

He advises caution when interpreting messaging from major tech companies, reminding us that "all of those messages are sales pitches for a product that you're going to have to pay for."

Societal Impacts and Concerns

Vanguard addresses two major concerns about the societal impact of AI:

The "Useless Human" Class

While acknowledging the potential for AI to exacerbate existing social issues, Vanguard argues that the concept of a "useless class" is not new:

"We've always had that class. We have it today, we've had it a hundred years ago, we had it a thousand years ago, and we will have it in 100 years from today."

He traces the roots of this issue back to the Industrial Revolution, noting that it primarily affects manual laborers and those without specialized skills.

Education as a Safeguard

Vanguard emphasizes the critical role of education in protecting individuals from becoming obsolete in an AI-driven economy:

"The one single thing that you, dear audience, can do to not get exited out the bottom is education. If you are an educated person, if you have skills that go beyond manual labor, you will very unlikely be a member of this group."

Job Creation and Displacement

While acknowledging that AI will create new jobs, Vanguard is realistic about the overall impact:

"Will it create as many new jobs as it has obfuscated the old jobs? Probably not. And will those be the same people? Definitely not."

He notes that the skills required for AI-related jobs often don't match those of displaced workers, creating a mismatch in the labor market.

Business Pressures and AI Adoption

Vanguard discusses the pressure felt by companies to deliver AI products and solutions:

Self-Created Pressure

"The pressure is mainly self-created. Many of the AIs that have been recently released, especially around this generative AI framework like ChatGPT, they're not really ready for deployment."

He cautions against rushing to implement AI solutions without clear use cases:

"A lot of C-level people are searching desperately for use cases and can't find them all that well, and that's a telling sign."

Advice for Businesses

Vanguard offers pragmatic advice for companies considering AI adoption:

"If it doesn't occur to you immediately what you can use these models for, they're probably not that much use for you, and that may not be a bad thing. Don't overdo it, especially at this point when these models are still a bit rickety."

Conclusion

Patrick Vanguard's insights provide a grounded perspective on the current state of AI, particularly language models. While acknowledging the revolutionary nature of recent advancements, he emphasizes the need for critical thinking, human oversight, and realistic expectations.

Key takeaways include:

1. AI language models produce impressive output but often lack factual accuracy and logical reasoning.
2. Current AI is best used for generating first drafts, with human editing required.
3. The primary threat from AI is its potential use for large-scale misinformation and propaganda.
4. Education remains crucial for individuals to remain relevant in an AI-driven economy.
5. Businesses should be cautious about adopting AI solutions without clear use cases.

As AI continues to evolve, it's essential to maintain a balanced view of its capabilities and limitations. By doing so, we can harness its potential while mitigating risks and ensuring that human intelligence remains at the forefront of decision-making and critical thinking.

Article created from: https://www.youtube.com/watch?v=ixyrOmeIJz0