AI winters have been characterized by waning interest and investment in the field. With the current explosion in AI capabilities, are we headed towards another winter, or have we learned from past experiences to avoid a complete freeze? What presents the greatest threat for another AI winter?
To understand if Artificial Intelligence (AI) will enter an AI Winter, we must define what an AI Winter is, as well as when and why it happened. There are presently significant risks to AI that can cause another winter. However, AI has incredibly good reasons to be hopeful about the future and avoiding another AI winter.
According to the Gartner hype cycle, an AI winter occurs when the technology fails to meet expectations and passes through the “peak of inflated expectations.” At this point, the technology enters the “trough of disillusionment,” where governments and companies re-evaluate their current funding levels and either reduce or eliminate the funding for those projects. Hopefully, the technology can survive this challenging time and later achieve value, eventually reaching the “plateau of productivity.” [1]
It is broadly recognized that there were two major AI winters: 1974 to 1980 and 1987 to 2000. [2] While there are multiple factors for the winters, funding was the principal reason in both AI winters. There are two sources of AI funding: public and private sectors. The AI public sector funding was reduced or eliminated in the 1970s when US DARPA and the UK did not observe the anticipated benefits of AI. This is very well illustrated in the 1973 Lighthill Report which documented that researchers failed to demonstrate how AI could answer the highly complex/mathematically challenging real-world problems. As a result of the report, the UK government ended funding for AI research. [3] Private funding was greatly impacted by recessions and severe stock market turn downs. The 1970s was a period of stagflation where companies were not making major investments in R&D. In 1987, there was a major stock market crash which severely impacted on the willingness of companies to invest in emerging technologies. When companies recovered from the 1987 crash, they did not re-invest in AI but rather chose to invest in improving collaboration through distributed computing like client/server topologies and networking technologies.
What are the current risks that could create a new AI winter? I see several factors that can cause another AI winter: regulatory impacts, AI trust/ethics, security/privacy issues, chip/compute limitations and environmental concerns like power consumption and cooling. Regulatory presents the biggest risk. Governments that are too restrictive in their regulations can stifle competition, creativity, and increase operating costs. For example, the EU AI Act took effect 1 August 2024 and is regulated by the European Commission. Besides impacting speed to market and creativity, the cost of compliance will be impactful and is estimated between 1% to 2.7% of revenue. [4]
AI trust/ethics is another risk where governments, companies, and individuals can severely impact the whole AI marketplace by unethical acts. Security/privacy concerns are a major challenge. AI must protect the individual privacy and ensure the necessary level of security to protect their information. A massive breach will cause a loss of confidence in AI and slow the adoption of technology.
Chip/compute limitation is a concern. AI requires significant computational power and companies like Nvidia, AMD, Google, and Amazon need to continue the development of high performing chip sets to manage the demand. Lastly, there are the environmental concerns of power consumption and cooling. The previously mentioned compute platforms consume considerable power and require significant cooling systems. Data centers can consume over a gigawatt of power, more than entire towns require. In today’s world of ESG, fossil fuel/nuclear power is not extremely popular and green energy is currently inadequate. Data centers also require advanced cooling systems that utilize vast amounts of water, which then need to be discharged. These risks, while concerning, can be mitigated.
The current AI accelerators, I believe, are going to keep AI from going into another winter. These include the democratization of AI, adoption rates, private sector funding, chip development and quantum computing, AI Software platforms/re-useability, and Cloud availability.
The biggest AI accelerator is the democratization of AI. AI is now ubiquitous. It is on all our devices and the ease at which we can access AI will ensure that AI does not go into another winter. Everyday people, students, parents, and employees, all have access to AI. Related to democratization, the AI adoption rate is a deterrent to another AI winter. Different AI solution types have already been implemented by 48% of companies, and 45% of companies are expanding their investment because of the ChatGPT rollout in 2022. [5]
Unlike the 1960s, 1970s, and 1980s, the current principal AI funding source is the private sector. The hyperscalers (i.e. AWS, GCP, Oracle, Meta, Microsoft) alone are investing billions of dollars in AI. This does not include all the enterprises making comparable AI investments.
As a risk, the chip/compute limitation is a concern. However, companies like Nvidia, AMD and others are making large investments and advancements in building the next generation of chips. In addition, Quantum computing is making considerable progress and ensures that we will have the necessary compute platforms into the foreseeable future. The AI software platforms like Python are making programing much easier and are addressing re-useability by providing common/open-sourced software libraries to reference previously written software and algorithms. This makes AI applications less expensive to generate and maintain.
Lastly, the advent of Cloud is providing scale and cost efficiency for all levels of users, developers, and companies. SMB companies can access the compute resources which were previously limited to the large companies.
While another AI winter is possible, I do not foresee it happening. AI has been around since the 1940s and has clearly demonstrated its benefits to the public. We are at a tipping point where all the adjacent technologies like compute, storage, networking, cloud, software, and security are further enabling AI adoption by governments, business sectors, and individuals. Even though an AI winter may not happen, there will still be hurdles to overcome in further AI adoptions.
References:
- Jaffri, Afraz. “Explore Beyond GenAI on the 2024 Hype Cycle for Artificial Intelligence.” Gartner https://www.gartner.com/en/articles/hype-cycle-for-artificial-intelligence
- Wikipedia, “AI Winter.” Wikipedia. https://en.wikipedia.org/wiki/AI_winter
- Wikipedia, “Lighthill Report.” Wikipedia. https://en.wikipedia.org/wiki/Lighthill_report
- Microsoft, “The EU’s AI Act Explained” Bing Videos https://www.bing.com/videos/riverview/relatedvideo?q=eu+ai+act+pdf&mid=77E5D84565446C322C9277E5D84565446C322C92&FORM=VIRE
- Howard, Chris – Global Chief of Research & Distinguished VP Analyst, Gartner Research & Advisory, “Gartner’s Top 10 Tech Trends for 2024 | Full Keynote from #GartnerSym” Bing Videos https://www.bing.com/videos/riverview/relatedvideo?&q=2024+gartner+ai+hype+cycle+analysis&&mid=FF1C7971BD24521D2131FF1C7971BD24521D2131&&FORM=VRDGAR
A condensed version of this article is published on our Blog under the title Is AI Being Overhyped?
