Our friends over at Sony AI have prepared a special set of compelling technology predictions for the year ahead. The Sony AI team is comprised of researchers and leaders with backgrounds in deep reinforcement learning, data science, law, privacy and security, and more. They each offer different perspectives on topics related to AI ethics and policy, the use of AI to augment creativity and scientific research, emerging AI training methods, and more. From the company’s point of view 2024 should be quite a year! Enjoy these special perspectives from one of our industry’s best known movers and shakers.
AI Ethics Evolves “Beyond Technical Problem,” Now Recognized as Complex Socio-Technical Challenge
“AI ethics is not just a topic for technologists; it is fundamentally a socio-technical pursuit. The issues AI ethics address are complex and multifaceted because AI systems can reinforce biases, impact privacy, and influence decision-making, among other things. That means they cannot be adequately understood or managed from a single disciplinary perspective as they encompass technical, ethical, legal, social, and policy dimensions. In 2024, organizations that want to create ethical AI systems need folks who deeply understand the inner workings of AI algorithms and systems and how to modify the technologies. They will also need individuals who can understand more of the societal components, whether law, policy, social science, or ethics. I believe that collaboration among these experts can lead to more holistic and effective solutions. Because of this, we will see more and more organizations in 2024 build their own teams of experts in artificial intelligence, data science, computer vision, law, privacy, and more.” — Alice Xiang, Global Head of AI Ethics, Sony Group Corporation, and Lead Research Scientist, Sony AI
Growing Focus on AI Ethics Will be Adopted as a Cultural Mindset
“In 2024, organizations that want to implement AI ethics well and make it a significant comparative advantage of their products and services must have it embedded deeply within their products through AI ethics by design. This is only possible when there is alignment within the organization from the top to the bottom. Today, companies with robust, mandatory AI ethics assessment processes can do so because of internal alignment. Not only is it crucial to deploy fair, transparent, and accountable AI systems, but it shows that they are taking AI ethics seriously and not just using it as a way to try to promote products. Organizations must shift to this practice as there are increasing calls for more ethical AI and assurances that AI will not harm society in 2024 and beyond.” — Alice Xiang, Global Head of AI Ethics, Sony Group Corporation, and Lead Research Scientist, Sony AI
AI-Generated Content Will Prompt Greater Ethical Considerations Globally
“Ethical considerations regarding AI-generated content and the necessity for safeguards and regulations will be at the forefront of discussions in 2024. The global landscape will witness the rise of AI regulation, including the European Union AI Act and regulations in other regions like the USA and China, marking a significant shift towards responsible and accountable AI deployment. However, it’s crucial that amidst these discussions, we continue to emphasize AI’s potential for a positive impact on society, focusing on responsible deployment and equitable outcomes.” — Tarek Besold, Senior Research Scientist, Sony AI
Responsible AI Development Will No Longer Be an Afterthought, But a Critical Priority
“The proliferation of artificial intelligence in our daily lives – from the technologies we use at home, at work, and in the virtual world – will only continue in the years to come. However, as the race to bring these AI technologies to market intensifies, the technology industry must truly think about their research and development in a responsible way. These technologies must be human-centric, and we must fundamentally understand how they will impact human lives in every area they touch. In 2024, we will see deeper discussions on responsible AI development as well as critical work from the AI research community around benchmarks and tools that can be leveraged to ensure technologies are human-centric, fair, and accountable.” — Michael Spranger, President, Sony AI
The Identification of Ethical Data Collection Practices and Solutions Will Be Top Priority for Research Organizations
“Ethical data collection will only gain in importance given the rise of more general purpose AI systems that rely on huge amounts of training data to achieve their capabilities. There is growing awareness that many of the ethical problems in AI systems stem from issues with their training data. Still, there has been a lack of solutions to enable more responsible data curation. In 2024, we’ll see research organizations lead the charge in identifying solutions and new practices to address these issues – all with the ultimate goal of enabling the adoption of more ethical data collection practices that prioritize fairness, consent, privacy and IP protections, transparency documentation, and responsible sourcing.” — Alice Xiang, Global Head of AI Ethics, Sony Group Corporation, and Lead Research Scientist, Sony AI
Closely-Watched AI Policy Efforts Will Focus on Concrete Guidance and Enforcement
“2024 policy efforts will continue to focus on turning established AI principles into more concrete guidance for the industry on how to evaluate, develop, and deploy trustworthy AI systems. Furthermore, as we wait for the formal adoption of the long-anticipated AI Act, we should expect other jurisdictions to look to the EU to inform their own AI policy approaches. If the AI Act is formally adopted in 2024, EU member states will shift attention to enforcement. This transition period will be watched closely by governments that are considering their own risk-based regulatory frameworks for AI. While the EU is likely to have a Brussels Effect on global AI policy, there are likely to be divergent approaches in key AI jurisdictions like China that are focused on more use-case-specific and sectoral regulation. Due to resulting interoperability challenges, multilateral initiatives that can harmonize international policies and standards will continue to be a priority in 2024.” — Victoria Matthews, AI Policy Specialist, Sony AI
Collaboration Will Be Key to AI Policy Creation
“In 2024, policymakers and regulators must continue to seek AI expertise from the private sector, academia, and civil society – especially from public interest and civil rights advocacy groups. Collaboration of this sort is needed to ensure regulation is proportionate and enforceable, considers new and emerging AI risks, and adequately addresses safety, transparency, fairness, and accountability concerns. For example, proposed regulation would impose reporting requirements and/or independent evaluations on businesses for certain AI systems. This could help organizations identify risks, including potential biases, earlier in the AI lifecycle, and promote a degree of transparency about AI capabilities. However, feedback must be solicited from a spectrum of affected stakeholders to ensure such regulation effectively balances privacy and confidentiality concerns.
2023 has also heralded cooperation on an international level. Efforts to harmonize AI approaches have been bolstered by global agreements like the Bletchley Declaration and the G7 International Guiding Principles and Code of Conduct on AI. It can also be seen in the continued work of the Global Partnership for Artificial Intelligence (GPAI), the Organisation for Economic Co-operation and Development (OECD), the United Nations Educational, Scientific and Cultural Organization (UNESCO), and other organizations. In 2024, it is crucial that multilateral AI forums seek equitable inclusion from perspectives that do not dominate the current AI discourse, such as from countries in the Global South, where AI development brings unique opportunities, but where associated harms — including exploitative data practices — are often felt more severely.” — Victoria Matthews, AI Policy Specialist, Sony AI
AI Will Fuel a New World of Possibilities for Creators
“This is a very exciting time for artificial intelligence, especially for people like me who have worked in the field for decades. There has never before been a greater awareness and broad reach of artificial intelligence; people don’t have to be technical experts trained in computer science to use it. However, one of the most exciting developments is the opportunity it provides creators. Artificial intelligence can now be used as a helpful tool to creators, empowering them to magnify their ideas and work as well as uncover new possibilities. In 2024, I think we will see greater exploration of artificial intelligence among creators, and new opportunities and possibilities in creativity and imagination will be revealed.” — Peter Stone, Executive Director, Sony AI America
Pushing the Boundaries of AI Into Multisensory Experiences
“In 2024, several prominent topics are expected to dominate discussions in the field of AI. Among these, the continued advancement of generative AI will be a key focus. With AI systems becoming increasingly proficient at generating diverse content, including language, images, sound, and music, the discussion will become: can we push this further and create integrated and coherent multisensory experiences? For example, can we have systems that generate text, images, and music which naturally fit together, or can we even evolve them to include other modalities, such as smell or touch? And, if that works, how can we create sequences of content which provide truly immersive experiences for users?” — Tarek Besold, Senior Research Scientist, Sony AI
AI Will Become a Catalyst for a New Age of Innovation and Creation
“Artificial intelligence is no longer just a powerful tool for technologists and data scientists to deploy in industries such as manufacturing, logistics, or marketing. With recent advancements, it has become a catalyst for a new age of innovation and creativity as a much larger cohort of people have access to the power of this technology regardless of their background and experience with it. Creators around the world – from contemporary artists to storytellers, to game designers, to scientists – now all have artificial intelligence at their fingertips, empowering them to extend the boundaries of their imagination and creativity. We will begin to see the games, movies and music produced by the early adopters in these areas in 2024, showcasing the creation of tangible concepts that were once deemed impossible. This will inspire an entirely new generation of creators, fueling their curiosity for solving challenges, uncovering new experiences or modes of expression, or making the impossible possible.” — Michael Spranger, President, Sony AI
Revolutionizing Science: AI’s Power Unlocks an Uncharted Scientific Frontier
“Unlike the internet, which serves as a distributed storage device, AI’s conversational nature enables it to navigate uncharted territories of scientific inquiry. Generative AI and other emerging AI tools and methods can be harnessed to innovate the way we develop Trend and modeling in science. This marriage of AI and scientific exploration can usher in a new era of discovery, where machine intelligence and human inquiry propel us into areas previously deemed inaccessible and lead to radical new theories and breakthroughs across all scientific disciplines: medicine, biology, physics, and more. One of the key contributions of AI to scientific discovery lies in its ability to avoid human biases inherent in (traditional) research. By leveraging advanced algorithms and machine learning, AI can identify and eliminate biases that may permeate scientific investigations. This enhances the integrity of research findings and opens the door to exploring previously overlooked facets of various scientific disciplines. In the next year and many others to come, we will see scientists, researchers, and medical professionals increase their use of AI tools to further their work, offering unprecedented opportunities for innovation and breakthroughs.” — Michael Spranger, President, Sony AI
Advancements in Fast Generation Will Make Generative AIs More Interactive
“Diffusion models face challenges in slow sampling due to their reliance on multiple sampling steps, hindering real-time and interactive applications. Recent research has focused on distillation techniques for diffusion, achieving significant progress in speeding up sampling to just one or a few steps. Additionally, Generative Adversarial Networks (GANs) have demonstrated comparable performance to diffusion with only one step, thanks to extensive studies exploring diverse architectures trained on large datasets. This accelerated sampling speed will open up new possibilities in 2024 across gaming, location-based entertainment, and digital healthcare, enabling generation optimization based on surrounding conditions and much more.” — Yuki Mitsufuji, Lead Research Scientist, Sony AI
Human-in-the-Loop Training Methods Will Become Increasingly Popular
“Recently, we have seen developments in generative AI, generative adversarial networks, and diffusion models. However, I think one of the things that is also happening is the growing recognition of human-in-the-loop training methods and the possibilities they afford. These methods are different because it’s not just simply about the computer training itself from large quantities of data. Human-in-the-loop provides the computer with the opportunity to learn from human feedback and input that are given through demonstrations; evaluations, where a human indicates whether the computer did a good job or a bad job; and interventions, where a person watches how the program is doing and gives corrective actions when it does something that the person doesn’t want it to do. In 2024 and the next few years ahead, the class of human-in-the-loop training methods will become increasingly mature and more widely used, opening up a plethora of new possibilities for artificial intelligence.” — Peter Stone, Executive Director, Sony AI America
The Success of Reinforcement Learning as a Key AI Training Method Will Grow
“In 2024 and the coming years, we will see more and more examples of reinforcement learning as a successful training method in artificial intelligence. This is something we have been focused on at Sony AI as part of our Game AI work, and it has proven to be an outstanding method. But, in much of the rest of the world, there has been a greater focus on supervised learning and self-supervised learning methods – which can be attributed to some of the more recent public successes in artificial intelligence-driven by the maturity of software packages and toolkits. As reinforcement learning tools mature, I expect to see more widespread adoption.” — Peter Stone, Executive Director, Sony AI America
AI Training Methods: Moving Toward More Specialized AI Systems
“In the near future, AI training methods are likely to evolve in several ways. One notable trend is the move towards smaller, more specialized datasets. A lot of what has happened in AI over the last decade was enabled through the abundance of data available indiscriminately on the Internet. What we are now looking at is trying to build systems that have to be super precise and super-specialized. The challenge with creating these more specialized systems is that there is very little data to work with. So, while large datasets have been a driving force in AI advancement to date, we’ll see an increasing focus on making AI systems excel with limited data. This will require the incorporation of domain knowledge infused with machine learning to drive the next wave of AI progress, giving rise to neurosymbolic systems.” — Tarek Besold, Senior Research Scientist, Sony AI
Neurosymbolic AI Will Emerge as a New Paradigm, Showing Close Alignment with Human Reasoning
“In the coming years, I believe AI will undergo a significant transformation. Machine learning, which has been the dominant paradigm for the past decade, will evolve by merging with techniques from earlier AI approaches. Specifically, we will see significant progress towards the fusion of symbolic representations, knowledge semantics, and structured data with statistical machine learning. Currently, many AI models, including large language models, heavily rely on statistical patterns in data but struggle to tap into structured knowledge sources like the semantic web. In the future, we anticipate that AI will increasingly leverage this explicit knowledge to guide and sometimes constrain the statistical learning process. This emerging paradigm, often referred to as Neurosymbolic AI, combines neural networks with symbolic representations, creating a new approach that might align more closely with what we know about how human cognition processes information than most current approaches do.” — Tarek Besold, Senior Research Scientist, Sony AI
The Future of AI Training Methods: Self-Supervised Learning, Transfer Learning, and Federated Learning
“In the near future, several AI training methods will gain prominence and importance. Firstly, self-supervised learning will be a fundamental technology used to build large foundation models across various domains, including language, vision, and music. This method will be essential as most real-life data is unlabeled, and manual labeling is costly and uncertain in terms of quality. Self-supervised learning will empower model training without the need for extensive human annotations. Secondly, transfer learning will continue to be highly relevant, especially in conjunction with self-supervised learning. Large pre-trained models will be easily adapted to various downstream tasks in computer vision (CV), natural language processing (NLP), and more, leveraging the advancements in transfer learning techniques. Lastly, as privacy concerns grow rapidly, federated learning will become increasingly vital in the coming years. Federated learning offers a privacy-conscious learning approach, enabling data parties to retain control of their data. Instead of sharing raw data, these parties only need to share learned model parameters with a server, ensuring both performance and privacy.” — Lingjuan Lyu, Head of Privacy-Preserving Machine Learning (PPML), Sony AI
Sign up for the free insideBIGDATA newsletter.
Join us on Twitter: https://twitter.com/InsideBigData1
Join us on LinkedIn: https://www.linkedin.com/company/insidebigdata/
Join us on Facebook: https://www.facebook.com/insideBIGDATANOW
Speak Your Mind