Enormous global interest in AI is abundantly obvious and huge AI investments among giant IT companies are also widely known. Yet there are also many people, including prominent academics and researchers, who think AI is overhyped. While AI excels at pattern matching and producing lucid text, they say, it lacks genuine understating and makes too many user errors.
These gaps are true. Nonetheless, AI is highly impressive and has prodded many technology giants and venture capital-backed startups to invest hundreds of billions in dollars in science. They all believe AI continues to have no place to go but up, a realistic scenario. Meanwhile, more than 50% of Americans in the workforce already use generative AI to some extent.
Consider the surge in AI investment. According to Stratview Research, the market for graphics processing units (GPUs) – the heart of AI technology – soared to a sizzling $107 billion in 2024 – up from $15 billion in 2023. Stratview stated the market will grow to nearly $122 billion this year and increase further to $295 billion in 2032.
This doesn’t mean critics are off the rails. Despite incredible attention, AI faces multiple challenges spanning technical, ethical, and societal domains. One problem is that AI models, especially large language models, require immense amounts of high-quality, diverse, and representative data for training. This leads to inaccuracies and inconsistencies, producing flawed AI outputs deemed “garbage in, garbage out.”
Other issues include the fact that many advanced models are complex "black boxes,” making it difficult to understand why AI reaches a certain conclusion, sometimes undermining trust. Other problems include job displacement, already impacting major companies and discrimination and fairness. Specifically, AI systems can perpetuate and amplify existing societal biases, leading to discriminatory or unfair outcomes in areas such as hiring and lending.
Still other issues include a lack of skilled AI talent and the belief in some quarters that major companies are spending so much on AI that the enormous cost of GPUs may never be earned back.
In the case of skilled talent shortages—an issue for years—it's expected to get even worse as AI adoption accelerates across various industries. As for the struggle to put AI to good use financially, Silicon Valley venture capital firm Sequoia has said that the AI industry spent $50 billion in GPUs in 2023, but brought in only $3 billion in revenue, the latest figure available. This is deemed the “productively paradox” and persists.
Here are additional details about several challenges:
+ Job Displacement. Some employers are already curbing hiring in anticipation that AI can do the work instead. Among the most vehement is CEO Tobi Lutke, the leader of the Canadian e-commerce company, Shopify, who recently told his managers they cannot make new hires unless they can prove AI is incapable of doing the job.
Longer term, more company leaders are warning that AI efficiency could shrink corporate workforces altogether. Last month, Ford Motor CEO Jim Farley predicted at the Aspen Ideas Festival that AI will halve the number of white-collar jobs in America. Also in June, Amazon CEO Andy Jassey said the company’s workforce will shrink due to AI. In addition, Adobe CEO Shantanu Narayen recently said he isn't looking to grow his headcount dramatically because AI will pick up the slack.
Meanwhile, at Splunk, an IT security company acquired last year by Cisco Systems, an employment executive said every new job has to be approved at the C-Suite level -- again reflecting the performance power of AI.
+ Hallucinations in Large Language Models. This mostly refers to LLMs that generate seemingly confident output that is factually incorrect or entirely fabricated and not supported by its training data. It’s called "hallucination" because it’s creating information that isn’t truly there, similar to how a human might hallucinate. When chatbots confidently present false information as fact, they can spread misinformation or disinformation. Moreover, if users cannot consistently rely on information provided by a chatbot, they can lose trust in the technology, hindering adoption in sensitive domains such as healthcare.
+ Bias, Fairness, and Discrimination. AI systems learn from the data they collect. If this data reflects existing societal biases, regarding improper takeaways in racial, gender, and socioeconomic issues, AI can learn and perpetuate these biases. This can lead to discriminatory outcomes in areas such as hiring, loan applications, and criminal justice.
+ Artificial General Intelligence (AGI). This is the holy grail for many AI researchers. It refers to a hypothetical AI that possesses human-like intelligence across a wide range of cognitive tasks, fundamental to human learning and interaction. It would be able to learn, reason, understand, and adapt to new situations like a human, without needing to be specifically trained for every single task.
AGI is being researched by some of the most well-funded labs in the world as well as by numerous universities, including Stanford, MIT, and Carnegie Mellon along with independent researchers. The downside of AGI, as well as the upside, is hotly debated. The upside could potentially solve scientific challenges and eradicate many diseases. The downside, on the other hand, is that AGI could surpass human intelligence in all domains and become uncontrollable and misaligned with human values.
In addition, AGI progress could be capable of performing virtually any intellectual task better and faster than humans. This could lead to unprecedented levels of broad job displacement, far beyond what less advanced AI today is causing.
Notwithstanding its challenges, AI’s future prospects are vast and informative and eventually are expected to enhance the economy and many facets of life. Beware, however. AI must learn to be managed extremely well, not the case today. Otherwise, negative developments could conceivably overcome its allure.