Artificial intelligence (AI) and machine learning (ML) based language processing models were the hot topics of 2023. Boosted by the novelty of tools like ChatGPT, Midjourney, Soundful, and others, the public hype made it feel as if AI is radically changing everything we do.
Generative AI & the future of data centers: Part I - The Models
A seven-part article on what large language models and what the next wave of workloads mean for compute, networking, and data center design
However, at the end of the year, there are increasingly more signs that we are not quite there yet. Limitations of these tools, as well as legal and ethical challenges to their development, leave less to celebrate and more to do before another breakthrough in AI adoption.
Demanding results from AI
According to McKinsey's The State of AI 2022 report, AI adoption has settled between 50 percent and 60 percent over recent years and even went slightly down since 2019. Around the same time, multiple rollouts of tools based on generative AI made it seem that, on the contrary, we are experiencing a giant surge in AI adoption in business.
However, the scale of it was probably blown out of proportion. Companies like Salesforce and Microsoft race to be the world's first to introduce generative AI tools for particular tasks like summarizing customer data and generating real-time tips for meetings. Yet, even among businesses with $1 billion in yearly revenue, 60 percent are still a year or two away from implementing their first generative AI solution.
As long as the novelty and hype worked for good publicity, there was a reason to implement AI tools without expecting a prompt return on investment. That time has passed. In today's economic conditions, boards and investors will increasingly demand proof of positive results when authorizing AI adoption. So far, it seems that, at their current stage of development, the value generative AI tools can produce is limited.
How can AI/ML move forward?
Analysts from CSS Insight and Gartner find generative AI "overhyped" and predict that it will fade away from public interest already in 2024. Before AI can live up to this year's hype, we need to address the lack of reliability and accuracy that comes with superficially generating results according to statistical probability.
On the other hand, generative AI is just part of AI research. Moving forward, we might see the focus shifting from generative to causal AI and more nuanced machine learning techniques, such as federated learning.
While generative AI equates correlation with causation, advanced causal AI should function more like the human mind. It goes beyond statistics when examining the possible relationships between cause and effect. Thus, it can better discover what gives meaning to word sequences and produce more reliable results.
Federated machine learning is a framework in which ML algorithms can be trained without direct access to users' private data. In this decentralized paradigm, multiple partners with separate datasets train the algorithm collaboratively but without ever exchanging or pooling input data. This method can help solve the pressing issues of data privacy and isolated data islands.
This is essential technological innovation since accumulating legal cases regarding the privacy and ownership of data used to train AI already pose challenges to wider AI adoption. Courts and regulatory bodies agreeing on clear rules for further AI development and usage should also play an important part in addressing these challenges.
The market for generative AI will still grow
Of course, the Gen AI market is not going to roll back even if the general public will not watch it as enthusiastically as this year. The ML market is estimated to grow at 18.73% annually between 2023 and 2030, resulting in a market volume of $528 billion by 2030. We might even see new major players in the field of large language models (LLMs), providing training services and computing resources.
Gen AI is already making an impact on a number of industries, including marketing, design, and cybersecurity. The coming years might see it spreading into pharmaceutical, manufacturing, engineering, automotive, aerospace, and energy industries, maybe even streamlining core business processes.
The ability of businesses to adopt and deploy Gen AI further will depend on the providers' ability to serve these models as web-based APIs. Companies already implement ChatGPT into their daily tasks, such as customer care chatbots, generating leads, collecting product feedback, or summarizing video content. Learning the concept of causation and providing API access might allow Gen AI to be used in "harder" technical areas, like predictive maintenance.
In conclusion: the redefining year
To sum up, 2024 is going to be the year when we redefine the field of AI. After a long time of asking what AI could do, we are focusing more on what it should be enabled to do. Case law and national as well as intergovernmental institutions must provide some boundaries here.
Meanwhile, the market demand for quality over fast adoption should drive commercial AI developers to explore new areas. In all likelihood, Gen AI will not go away, but the field is going to be redefined by those striving for more intricate solutions.
