• Adi Andrei - Director at Technosophics, Ali Chaudhry - Founder at Veracious and Generative AI and RL Community in London, Juras Juršėnas - Chief Operating Officer at Oxylabs


• 05.12.2023 11:30 am
#AI #GenAI

2023 has seen an unprecedented spike of news and discussions around artificial intelligence (AI) and machine learning (ML) — after decades of research in closed scientific circles, AI- and ML-driven technologies suddenly made their way among the most desirable consumer-facing solutions. Today, AI tools are used by millions of people for getting information and doing textual tasks (ChatGPT), creating images (Midjourney, Stable Diffusion, DALL-E), writing code (GitHub Copilot), and even making music (AIVA, Soundful).

Although it sounds like we are already living in the future, it is worth remembering that most of this breakthrough is happening around generative AI (Gen AI) based on large language models (LLMs) and other types of transformer models, which is only a fraction of AI research.

However, can Gen AI truly bring the seismic changes people expect? Adi Andrei and Ali Chaudhry, prominent AI experts and members of the AI/ML advisory board at Oxylabs, a leading web intelligence collection platform, and Juras Juršėnas, Oxylabs’ chief operating officer, kindly agreed to share their ideas about the possible state of AI in 2024.

AI vs ROI: companies will exercise more scrutiny in AI adoption

According to McKinsey’s The State of AI 2022 report, AI adoption has settled between 50% and 60% over recent years and even went slightly down since 2019. Although multiple rollouts of Gen AI-based solutions might have created an illusion of nearly universal AI adoption, businesses remain cautious, says Adi Andrei, director at Technosophics, former head of data at SpaceNK, and ex-senior data scientist at NASA and Unilever.

“The economic situation makes companies more pragmatic when adopting new AI and data analytics systems. Boardrooms need proof that these investments will increase the bottom line. A lot of money and effort has been poured into monetizing ChatGPT and similar Gen AI solutions, but the results are lacking.”

According to him, reasons might differ from company to company, but high adoption costs and questionable reliability due to AI still suffering from hallucinations are the most common deal breakers. Recent predictions from CSS Insight and Gartner went as far as calling generative AI “overhyped” and forecasting it might fade away from public interest already in 2024. Adi Andrei agrees these statements might ring true:

“After all, LLMs appear intelligent only on the surface — they are stochastic, meaning they don’t have reason and function on statistical probability of different words following one another. Such superficial intelligence is not always valuable and reliable, and the industry is waking up to reality.” 

Adi Andrei notes that some companies today are trying to quickly pivot towards the so-called “meaningful AI,” which could be the next logical step in the evolution of Gen AI. However, getting there is not that easy.

“It took many decades to get where we are today, and faster progress won’t be accomplished by just throwing in more computer power. It will require deeper thought and consideration to answer such questions as how do we create meaning and what makes something “meaningful”. Is meaning a subjective experience and, therefore, challenging for a machine to relate to? Or is there more to it? I am looking forward to these discussions,” said Adi.

Gen AI should impact different industries

Ali Chaudhry, founder at Veracious and Generative AI and RL Community in London, expects that Gen AI-powered applications will continue disrupting different sectors, including healthcare, education, financial services, and supply chain management.

“Gen AI is already affecting many industries, ranging from marketing and design to legal services. It should also proliferate into pharmaceutical, manufacturing, engineering, automotive, aerospace, and energy industries, streamlining and augmenting core business processes. Unfortunately, given the ongoing political situation across the globe, we might as well witness growing investments in AI applications in the defense sector.

“Although the economic and business climate feels predictably unpredictable, it is estimated that the ML market will grow at 18.73% annually between 2023 and 2030, resulting in a market volume of $528 billion by 2030. I strongly believe we might see new major players in the field of LLMs, providing training services and computing resources,” explains Ali Chaudhry.

According to him, next year, the industry might also see some progress in legal and institutional AI regulation, however, it is unlikely to bring concrete or legally binding norms yet. “Conversations on AI and data ethics are getting more intense and louder; as we could witness this year, some of them spill out to courtrooms. So at least broad-level agreements on what is and what is not proper conduct are indispensable, especially regarding the issues of data privacy, bias, and AI misuse for criminal activities,” points out Ali Chaudhry.

Juras Juršėnas: regulation around AI and web data collection might finally get moving

Juras Juršėnas, the chief operating officer at Oxylabs, agrees that next year will see an increased interest in legal and ethical questions of web data scraping and AI training. “Many AI systems, generative AI in particular, rely on ML technology that feeds on a massive influx of data to train underlying algorithms and maintain accuracy. Unfortunately, restrictions on public web data collection might delay innovations in the AI field. On the other hand, the web data collection industry has long lacked clear guidelines and answers regarding data ownership, privacy, and data aggregation at scale. So, we hope that case law will start clearing up those gray zones.”

According to him, there has been so much talk around Gen AI recently that many have forgotten it is just a part of AI research — there are other emerging techniques that might offer much-needed technological breakthroughs.

“Federated machine learning and causal AI might help create a healthy competition in the AI field, which is currently dominated by only superficially intelligent generative systems. Federated learning is a framework that allows training ML algorithms without direct access to users’ private data, at the same time solving the pressing issues of data privacy and isolated data islands.

“Causal AI, on the other hand, offers hope to solve the problem my colleague Adi mentioned when talking about stochastic predictive models that fall short of objectiveness and accuracy because they are equating correlation with causation. Causal AI functions more like the human mind, asking such questions as “what if” and examining the possible relationships between cause and effect,” explains Juras Juršėnas.

Finally, coming back to generative AI systems, Juras anticipates that wider deployment and adoption of this technology in business organizations and beyond will depend largely on the providers’ ability to serve these models as web-based APIs.

“Some companies have already implemented ChatGPT into their daily tasks, such as customer care chatbots, generating leads, collecting product feedback, or summarizing video content. Although these are “soft” tasks, it is expected that models as APIs will also foster AI deployment in technical areas, such as predictive maintenance,” concluded Juras Juršėnas.