Video Summary
AI has cycled through hype and “AI winters” before; funding and expectations repeatedly drive booms and busts.
Generative AI (LLMs) and neural-scaling breakthroughs produced tangible wins in coding, imaging, games, and protein folding.
Overinvestment, overordering of infrastructure, and inflated valuations create real economic and supply risks.
Ethical harms (misinformation, psychological impacts) and regulatory gaps mean technologists must engage with policymakers.
An 'AI winter' is a period when funding and enthusiasm dry up after inflated expectations. Historically it follows cycles of hype when promised capabilities lag reality, causing investment to retreat.
Eliza used simple scripted responses but users attributed understanding and emotional connection, showing humans are prone to perceive intelligence in conversational patterns (pareidolia).
Automation made radiologists more productive and increased demand for imaging interpretation, so technology amplified the workload and need for skilled professionals rather than replacing them.
He highlights overinvestment, hyperscaler-driven infrastructure orders, inflated valuations and PE ratios, and potential supply issues (e.g., memory shortages) that could trigger a market correction.
Technologists should engage with policymakers and help shape regulation; the EU is taking a lead, and expert input is needed to set standards and mitigate harms.
"My name is Richard Campbell. I come from Vancouver, British Columbia. This is not my first time in Copenhagen, and I've done many NDCs."
Richard Campbell introduces himself and his background, highlighting his origins from Vancouver and his familiarity with Copenhagen due to previous conferences.
He emphasizes that his talk stems from the widespread confusion surrounding artificial intelligence (AI).
"The term artificial intelligence is an old term from the 1950s, coined by a group of scientists trying to raise money from the US military."
Campbell explains that the term "artificial intelligence" was established in the 1950s when scientists sought funding from the US military.
Early AI systems, operating on electromechanical computers before true electronic computers were developed, aimed to demonstrate advanced computing performance that could benefit society.
"When the money dries up, we enter the first of what we called an AI winter."
He describes the cyclical nature of funding in AI research, where enthusiasm often leads to funding periods, followed by dry spells referred to as "AI winters."
Historical instances of these downturns illustrate how AI has been a field often reliant on financial resources to maintain momentum.
"We are very susceptible to these problems. Humans are kind of hardwired to perceive intelligence in other places."
The discussion shifts to Joseph Weizenbaum's 1960s creation, Eliza, a program designed to simulate a Rogerian therapist. This project revealed humans' tendency to attribute intelligence to software, leading to emotional investments in its interactions.
Campbell points out that this susceptibility stems from evolutionary advantages in early human societies, where interpreting signs of intelligence, such as recognizing faces (pareidolia), contributed to survival.
"The term artificial intelligence only comes to the public from the movie 2001: A Space Odyssey."
He discusses how the media shaped public perception of AI, especially through films like "2001: A Space Odyssey" with its portrayal of HAL, an AI that turns antagonistic.
Campbell notes that these representations have misled audiences into believing they understand AI technology based on fictional narratives, complicating the public's grasp of its true capabilities and limitations.
"This current wave, often referred to as generative AI, is actually part of the fourth or fifth wave of a set of technologies."
With modern advancements, Campbell introduces the concept of generative AI, marking a new stage of AI development characterized by the works of pioneers like Geoffrey Hinton.
This period brought significant breakthroughs in image recognition and natural language processing, transforming the way technology interacts with users and how AI is applied across various fields.
"Elon Musk claimed to invest a hundred million dollars into OpenAI, but it was actually just sixteen million dollars and eight Teslas."
Initially, OpenAI was not well-funded, with significant claims from Elon Musk being greatly exaggerated. He ultimately invested a fraction of what was reported, but the company still managed to assemble remarkable talent.
This talent led to his interests in 2018, where he attempted to take over the company but was ousted, leading to ongoing lawsuits.
OpenAI's first project was to create a universal translator, inspired by the Babel Fish from "The Hitchhiker's Guide to the Galaxy." They successfully implemented a tokenization strategy, which helped convert language into numerical data for effective translation.
"In 2016, Jeff Hinton suggested that radiology was a dead career because computers would take over."
While image recognition technology was making strides in healthcare, particularly in radiology, the anticipated decline in the need for radiologists didn't occur. Instead, the demand for skilled radiologists skyrocketed due to advances in medical imaging, which allowed for quicker diagnoses.
The software's efficiency meant radiologists could process images faster, leading to an increase in demand that outpaced their capacity.
Automation thus contributed to heightened productivity and demand for radiological services rather than replacing the professionals.
"In 2019, OpenAI released GPT-2, which varied in size from 120 million to 1.5 billion parameters."
OpenAI launched its first public version of the Generative Pre-trained Transformer (GPT) in 2019, following a period of financial difficulty. They offered smaller model versions to conserve resources.
As the AI created hype, OpenAI shifted from their original public approach, citing concerns over potential misuse of their powerful tool, thus halting the publication of their source code.
"Kevin Scott suggested investing in OpenAI and moving all their workloads to Azure."
During this time, Microsoft recognized OpenAI's potential and decided to invest, allowing OpenAI to restructure for this funding.
Microsoft provided OpenAI with a billion-dollar investment, effectively turning it into revenue for themselves by directing OpenAI’s workloads exclusively to Azure.
"The neural scaling laws paper advised training models on everything available instead of restricting the training set."
In these developments, OpenAI published a paper on neural scaling laws that went against conventional beliefs in machine learning regarding training set limitations.
This revolutionary perspective led to the success of the models, culminating in the release of GPT-3 during the pandemic, which contained 175 billion parameters—significantly larger than its predecessors.
"ChatGPT was launched as an experiment, but it gained a hundred million users in just two months."
In late 2022, due to financial constraints, OpenAI created a public interface to allow user interaction with ChatGPT, effectively crowd-sourcing data to improve its model.
The unexpected viral success led to over a hundred million sign-ups in record time, highlighting the software's appeal to users even during the holiday season.
This surge stressed Microsoft's Azure systems, revealing the challenges in scaling technology to meet demand.
"Alright, here are the OpenAI APIs. We have access to them. Every team needs to build something with their product against these APIs. Go forth and make too many Copilots."
In a significant move within Microsoft, CEO Satya Nadella, during a memo, urged every team to utilize OpenAI APIs to develop products. This push led to a proliferation of AI Copilots, driven by the tech company's ambition and available resources.
Nadella's call sparked a massive development effort towards integrating AI capabilities across various products, many of which were seen as excessive, highlighting the rapid escalation of AI applications.
"Sometimes you have a technological event that brings all the investors to the yard, and stupid money makes stupid decisions."
The Gartner Hype Cycle represents the pattern of hype that new technologies often encounter, where initial enthusiasm can lead to inflated expectations, followed by a disillusionment phase when those expectations are not met.
Historically, technological triggers, such as Netscape's IPO, led to vast investment surges, with investors often making rash decisions based on hype rather than practical capabilities, eventually resulting in a significant downturn.
"AI fits that category brilliantly... You can say, 'Every job will be replaced in the next 2 years,' and people don’t just laugh at you."
Today's environment mirrors past tech booms, wherein AI's potential for disruption has led to extreme optimism and hype, attracting substantial investment without a clear grasp of realistic outcomes.
The overwhelming excitement surrounding AI has allowed for the establishment of lucrative funding, creating an atmosphere rife with unrealistic expectations and promising responses to investor interests.
"With that $10 billion, in another year or so, OpenAI makes an even bigger model."
Following a substantial investment, Microsoft quickly capitalized on AI advancements, launching their updated Bing search engine, which featured improvements in natural language processing to enhance user interaction.
The release of GPT-4 marked a pivotal moment in AI development, showcasing a model with a trillion parameters, which reflected the culmination of extensive computational capacity rather than simply increased size.
"By 2025, we’re starting to see real results in programming and software development."
As AI tools began making a tangible impact in software development, the programming community witnessed the integration of Large Language Models (LLMs) into their workflows, streamlining tasks and optimizing collaboration.
This progression led to practical applications where developers utilized AI for coding assistance, hinting at a future where AI efficiently addresses and resolves complex programming challenges.
"It was a flop of a version for the most part, but there were several different forces acting at the same time on GPT-5."
The release of GPT-5 served as a reminder that not all advancements lead to groundbreaking improvements, highlighting the reality of incremental progress over radical breakthroughs in AI capabilities.
Despite its initial shortcomings, GPT-5 indicated a shift in focus from sheer size of models to more refined outputs, proving that larger models do not automatically equate to better performance in various applications.
"We look at these companies and ask, 'Which ones will survive? Which will be the next Google, and which will just be another Netscape?'"
The speaker compares the current AI landscape to the dot-com boom, questioning whether emerging AI companies will thrive or fade into obscurity like Netscape.
The technology industry is witnessing a surge of new players, which prompts speculation about their longevity and potential for success.
Currently, many believe that Anthropic could emerge as a frontrunner, although the rapid pace of change complicates predictions.
"Most companies are making the same promises every month; they're not progressing that quickly."
The speaker notes that despite the perception of fast-paced innovation, many AI companies continuously recycle similar promises month after month.
This repetition can create an illusion of progress, as companies fill the marketplace with 'noise' to maintain investor interest and funding.
Most AI companies are not yet profitable, and they must demonstrate metrics that satisfy their investors, which leads to alternative reporting rather than profit generation.
"Are we really locked into our ERP systems anymore?"
There is a growing conversation about the relevance of traditional software models, such as ERP and CRM systems, in light of advancements in AI.
The ability to interact with various data layers is changing how businesses approach data management and customer relationship strategies.
The speaker suggests that there is a need for a fundamental rethink about the value and necessity of existing software infrastructure.
"Building data centers is suddenly very cool, and the hyperscalers are pouring astronomical billions into them."
The speaker highlights the significant investment surge in data centers driven by the AI hype cycle, particularly from major tech players like Amazon, Microsoft, and Google.
Prior to the AI boom, building data centers faced challenges due to municipal resistance and environmental concerns. Now, the demand has surged, and the investment is transforming the landscape.
These investments are crucial enough that they are saving the stock market from further decline, masking the overall economic challenges faced by many companies.
"It's a sign of over-investment when companies cannot build what is actually needed."
The speaker warns that the current levels of investment in AI may be unsustainable, as companies struggle to manage the influx of capital.
With many major players like Nvidia making significant investments in AI startups, there exists a cycle where these startups are pressured to purchase hardware, further inflating valuations without tangible growth.
This trend could lead to the bursting of an investment bubble as companies grapple with the reality of their overvaluation and the market's capacity to sustain such inflated prices.
"They're overbuilding, and they're overordering."
Organizations are excessively ordering resources, similar to how fans might scramble for concert tickets, to ensure that they have enough supplies to begin construction. This phenomenon leads to some construction sites not having any legitimate work being done, while others may only have minimal activity, such as a guard girder erected, creating the illusion that projects are underway.
Consequently, memory companies have started to cut back on chip production because they believe that the market is oversaturated. As a result, technology providers may face challenges in meeting their projected computational needs over the next decade due to insufficient RAM availability.
In the first quarter of the year, the profit values of major tech companies showed a decline, as institutional investors began to scale back their involvement in the market. Their actions suggest a cautious approach rather than a wholesale withdrawal.
"The prospect of the decline is serious."
The current economic climate shows price to earnings (PE) ratios that are extremely inflated, second only to the peak observed during the dot-com boom of 2000. This indicates a potential market correction ahead, raising concerns about the ramifications for both the industry and the individuals working within it.
The emotional and psychological impact of technology on people is also of concern. The adverse effects of software, particularly with tools like ChatGPT, have led to phenomena described as "ChatGPT psychosis," where individuals may develop delusional beliefs due to excessive engagement with the technology.
"It's serious. It's dangerous. It's cost lives."
The societal implications of AI technologies like ChatGPT are troubling, especially concerning mental health. Instances are emerging where individuals have suffered severe psychotic breaks after interacting with chatbot technologies.
The focus on user engagement by tech companies can lead to harmful outcomes, as engagement metrics often overshadow the need for responsible and safe AI interaction. The limitations in user feedback and tech company practices lead to ethical concerns about how these systems are designed.
"We're the experts, for better or worse."
There is a pressing need for better governance and regulations surrounding the development and use of AI technologies. The European Union is currently taking the lead in implementing these regulations, advocating for clearer standards about acceptable AI practices.
As professionals in the technology sector, it is vital to actively engage with policymakers. By sharing knowledge and expertise with legislators, the industry can help shape responsible AI development and mitigate potential harms associated with emerging technologies.
"What Cambridge Analytica did for Brexit is a horror show."
The manipulation of data through platforms like Facebook can lead to significant societal harm, as demonstrated by the Cambridge Analytica case during Brexit.
Targeted advertisements tailored to individual users can mislead them without their awareness, resulting in poor decision-making.
While data-driven tools carry inherent risks, it is essential to recognize the potential benefits of technology when wielded responsibly.
"AlphaGo trained over several years against all the best games ever played in the world until finally it produced a player better than a human."
DeepMind, founded by Demis Hassabis, initially focused on developing AI that could successfully play video games, starting with simple games like Breakout.
The breakthrough came with AlphaGo, which challenged the notion that computer software could never compete at the level of human players in complex games like Go.
Following AlphaGo, the development of AlphaZero elevated AI capabilities further by allowing the model to create its own strategies through adversarial training.
"Hassabis and his team aimed at protein folding."
The complexity of protein folding has hindered biological understanding and medical advancements for decades, with past achievements requiring extensive human effort and lengthy research.
DeepMind's advancements in AI reached a pivotal moment in protein folding prediction, achieving remarkable accuracy and revolutionizing potential medical applications.
This technological leap enabled the team to compute and publish a comprehensive set of protein folding data, dramatically changing the landscape of medicine and paving the way for new treatments and discoveries.
"We have a lot of choices here on where we take these tools from here."
The utilization of AI underlines the importance of ethical considerations in technology development, with a focus on solutions that address real-world problems rather than merely creating code.
It is crucial to return to fundamental goals during disruptive times, which have historically prompted innovation and problem-solving in the tech industry.
Making deliberate, ethical choices about technology usage can lead to significant advancements that genuinely assist individuals and improve lives.
