Open Nav

The future of AI: Predictions from industry leaders

2022 has been a breakthrough year for AI. Despite difficult times for the technology sector, AI is evolving, advancing and flourishing. ML insider, cnvrg.io’s annual survey on the state of machine learning has found that 89% of organizations are seeing the benefits of their AI solutions in 2022. 

AI is becoming more embedded into businesses and integrated into day to day life. With so much potential, everyone is looking to the professionals to answer the big question – where is it going? What’s next? How will AI evolve in the next 5 years?  cnvrg.io asked some of the leading minds in AI to share their predictions and insights to give readers a glimpse into what the future of AI may hold.

For the full list of AI developers to watch in 2023 go here

Irene Solaiman, Policy Director, Hugging Face

In the next 5 years I expect two avenues of progress: incredible leaps in capability innovation and the increasing urgency to evaluate and mitigate AI harms. I’m eager for those two avenues to merge; AI risks must be evaluated and addressed through all stages of development. The critical considerations that should be examined from project proposal to deployment include harmful biases, disparate performance, and misuse. Since I’m most familiar with generative AI systems which are considered “general purpose”, I hope there will be and am working to make more tools to measure AI behavior. We must ensure AI enhances and does not overwrite societies and cultures.

Ria Cheruvu, AI Ethics Lead Architect, Intel Corporation

The next 5 years present a tremendous opportunity to develop and refine intelligent, responsible, and human-centered AI systems. The emergence of cognitive computing, AI for the Internet of Things, and new vertical applications for AI, such as physics-based modeling and personalized educational experiences, can help transform the way we think about productivity and quality of life!

Massimo Belloni, Data Science Manager, Bumble Inc

The next 5 years will continue to see successful breakthroughs in research – pushing the boundaries of what possible (eg. diffusion models and generative AI art, AlphaFold). I predict we will also notice a good understanding of what AI/ML “actually” is, with successful ML deployments at companies at every scale ( thanks to MLOps nowadays being largely commoditised by big cloud vendors!) to improve specific challenges they are facing and making their operations more efficient.

Alessandra Sala, Sr. Director of AI and Data Science at Shutterstock, and Global President of Women in AI

In the next 5 years I see our lives and decisions intrinsically influenced by the outcomes of AI. AI is disrupting every industry from transport to medicine to agriculture and more. The participation of women and minorities in the design and development of these systems is paramount to achieve equitable outcomes for everyone.

Petar Veličković, Staff Research Scientist, DeepMind

I imagine that the next 5 years will see a very strong push towards achieving artificial general intelligence (AGI) and the plethora of amazing products and scientific discoveries that AGI will enable. Specifically, I expect a ‘synergistic’ union of our existing state-of-the-art perceptual models, with the rapid development of models for robustifying neural reasoning.

Anna Kostikova, Director Data Science and Machine Learning, Novartis

I predict that in the next 5 years large language models will definitely become the new platform technologies and will drive the evolution of how end users interact & use AI in their daily work and life. We will likely see an explosion of marketplaces built on top of “centralised” LLMs with consumer facing apps that will further proliferate an adoption of the AI-driven tools into consumer and business behaviors.

Leonardo Neves, Manager, Applied Research, Grammarly

AI development in the next 5 years will continue to push the boundaries for how large models can get, but we will see a shift towards more efficient and explainable models. With more efficient models, more researchers outside the most well-funded labs and companies will be able to contribute to the advances of AI in a meaningful way. As more people have access to this technology and can better understand how models make their decisions, AI will be used in even more use cases and creative ways.

Delina Ivanova, Director, Analytics, Mistplay

In the next 5 years I’m most excited about everyday applications of AI, specifically the automation of decisions and processes within businesses. We often think of AI as self-driving cars and robots, but in reality AI solutions can be simple and practical, and implemented fairly quickly. As companies continue to invest in data teams, and as more people up-skill and enter the field, I think “smart technology” or “smart solutions” will become an every-day expectation in business. Overall, this effort to automate creates time and space to focus on growth and evolution as opposed to maintenance. I also think we will see meaningful growth in AI applications in physical industries, for example mining, construction, and manufacturing. The solutions that exist today could be expensive, difficult to implement, or too specific to a product, but focus and growth in this space – especially in highly personalized/custom-product offerings – will create more efficiencies and ability to share good practices across industries.

Gaurav Chakravorty, Software Engineering Lead, Video Recommendation, Meta

I see AI evolving in 2 ways: First, enhanced usability, predictability and transparency. Most backend developers will have the skills needed to adopt machine learning in their applications. Developers and product managers won’t have the I-dont-know-whats-really-happening hesitation. Second, more focus on industrial knowledge of how to use it, less on model complexity and beating humans. Essentially AI is moving from the evangelization stage to the pervasive utility stage.

Nermeen Louizi Ghoniem, AI Engineer, Jabra

AI has tremendous potential for good, which we absolutely must harness in the next 5 years and forward. We need to change the mindset and attitude engineers have when engaging and developing AI. What I mean here is I’d like to please ask engineers, including myself, to acknowledge that yes, we have biases. And, that how these algorithms are going to look at any group of people and their role in society, it mainly comes from their creators – which is us. AI cannot be naturally fair or for example, gender-neutral, and engineers should not think about these issues as something divorced from their own role. And associated with that is encouraging more fact-based conversations about how all of this works. We need to stop hyping up the field, as we are only clouding laypersons’ understanding of the field. This is pivotal as I earnestly believe AI and technology as a whole can help bridge the digital divide and create an inclusive society, if not within the next 5 years, then within the next 10 years. To empower and educate people, we need to reveal and make more transparent how these analyses are being done and how to interpret some of the outcomes and decisions. We need to involve all of us in shaping the future of technology and our digital society.

Bayan Bruss, Senior Director, Machine Learning Engineering, Capital One

There is so much to look forward to in the next 5 years. Some of the main advancements I can project are a unification of the ML stack, abstraction of DS activities through engineering, the incorporation of models as “agents” in a multi-agent system, and finally, fewer models requiring complex inheritance structures to adapt to a wide range of tasks.

Across industry we are seeing the environmental complexity in which ML models are deployed; comprising a variety of “agents” including, rules engines, auditors, and compliance officials. By and large, models are treated as software objects that need observation. The evolution over the next five years will create a system that allows each agent – whether human, rule, or model – a common language for communicating about its environment while also being aware of the other agents at play and adapting to work alongside them.

Most companies are building models for the same customers across a wide variety of data streams, with different targets depending on the decisions. This results in a large number of model pipelines with tremendous redundancy. Given that the underlying data generating process is the same (just captured in different ways), it makes sense to explore transfer learning, self-supervised pre-training, and other deep learning techniques to build a base layer model of customer behavior that can be adapted for specific usage. Combining this with the ability to interact with other ‘agents’ in the overall system will provide companies with new levels of intelligence and automation.

For the full list of AI developers to watch in 2023 go here

Top MLOps guides and news in your inbox every month