2023 saw several "black swan" events - a slowdown in the U.S. inflation, a rebound in public markets following their 2022 slump, and a liquidity crunch in private capital markets. Amidst these, artificial intelligence (AI) emerged as a focal tech trend. Reflecting on the past year, AI lived up to the hype, with tech-heavy Nasdaq surging 43%, nearly reaching its peak in 2021. OpenAI, the catalyst of the generative AI wave with ChatGPT, soared to a $100 billion valuation, ranking third globally among unicorns, just behind SpaceX.
As 2023 closes, what seismic shifts can we anticipate in tech for the year ahead? We compiled AI predictions from leading institutions and independent tech thought leaders for 2024, focusing on the 10 most frequently mentioned concepts and forecasts.
Without further ado, let's take a look!
1. Multimodal AI: Model Multimodality Will Be the Next Frontier.
"The interfaces of the world are multimodal. We want our models to see what we see and hear what we hear, and we want them to also generate content that appeals to more than one of our senses," said Mark Chen, head of frontiers research at OpenAI, in a November 2023 presentation at the conference EmTech MIT.
A consensus among key opinion leaders: in 2024, integrating multiple modalities into a single model, termed Multimodal AI, will lead. This mimics the human ability to process varied sensory data. We'll see native multimodal models merging text, image, audio, and video within a unified framework.
The real-world applications of multimodal AI are diverse and expanding. For instance, it can be applied in simulations of complex physical systems like augmented reality, virtual reality, and digital twins; it may even allow users "to query information regarding tables, charts or schematics with a combination of text, voice, and images for responses more relevant to the context". This also implies a growing demand for multimodal training data.
2. AI Agents: Autonomous Intelligent Agents Will See Rapid Growth.
"The next big thing in 2024 will be an explosion in AI agents of all kinds, focusing on every consumer need and on every kind of business transaction," said Gokul Rajaram, Product and Business Helper at DoorDash.
Intelligent agents - programs capable of autonomous, proactive decisions based on environment, user input, and experience - are believed to be the next hotspot in AI. OpenAI's co-founder Andrej Karpathy said at a public event that OpenAI is more focused on changes in the agent field than model training methods. Experts at NVIDIA predict that in 2024, Autonomous Intelligent Agents will become more accessible to deploy across virtually any platform.
"2023 was the year of being able to chat with an AI. In 2024, we'll see the ability of agents to get stuff done for you. Make reservations, plan a trip, connect to other services," wrote computer scientist Peter Norvig, a fellow at Stanford's Human-Centered AI Institute.
Combining agentic AI with multimodal AI opens new avenues. These agents, gathering more comprehensive and authentic data via various sensors, can perform a wider array of tasks with less human intervention. Fully autonomous agents are seen as artificial general intelligence where AI agents are perceptive without needing human guidance, but this is still theoretical for now rather than practical.
3. Shadow AI: Shadow AI May Bring Data Security Issues for Companies.
“CIOs have struggled with ‘shadow IT’ in the past and will now confront ‘shadow AI’ – solutions used by or developed within an organization without official sanction or monitoring by IT. Well-intentioned employees will continue to use generative AI tools to increase productivity,” said Jay Upchurch, Chief Information Officer, SAS
Experts forecast that 60% of employees will use their own AI software for work tasks, boosting productivity and expectedly increasing employee ARR by 10-15%. However, this raises concerns about "Shadow AI". Employees might use chatbots without IT vetting, lacking awareness of security, data privacy, and compliance, potentially exposing sensitive information.
As Terry Ray from Imperva warns, people don't need malicious intent to cause data breaches. In most cases, people just want to improve work efficiency. But if companies turn a blind eye to LLMs querying their backend code or sensitive data storage, things blowing up in their faces is only a matter of time.
4. RAG: Retrieval-Augmented Generation Will Spur Enterprise AI Adoption.
"From RAG to riches: Expect to hear a lot more about retrieval-augmented generation as enterprises embrace these AI frameworks in 2024.," said Kari Briski, Vice President of AI Software.
Generative AI tools still grapple with hallucination issues, posing obstacles to enterprise adoption. Retrieval-augmented generation (RAG) merges text generation with information retrieval, enhancing the accuracy and relevance of AI-generated content. Via semantic search, LLMs can access external information, so enterprises don't need to store all knowledge directly in the LLM, which also reduces the model size to improve speed and lower costs.
"You can use RAG to go gather a ton of unstructured information, documents, etc., and feed it into a model without having to fine-tune or custom-train a model," said Matt Barrington, Americas emerging technologies leader at EY.
This will undoubtedly boost enterprise AI adoption, as having the latest information is crucial for enterprise applications. By combining RAG and foundation models, enterprises can achieve more accurate generative AI applications in healthcare, finance, retail, manufacturing, and more, with fewer resources.
5. Transformer Alternatives: Non-Transformer-Based Models May Take Off.
"The next big thing in 2024 will be AI models that push beyond the limitations of transformer architecture, giving us larger context windows, faster and cheaper inference, and more powerful AI systems overall," said Bucky Moore, Partner at Kleiner Perkins.
Since Google's 2017 paper "Attention is All You Need", Transformer architectures have dominated AI – models like GPT-4 and Claude all adopt uniform Transformer architecture. Transformers essentially predict the next likely word in a sentence. However, such models have been criticized as "parroting randomly" instead of true reasoning. Moreover, their training and inference costs are staggeringly high.
Some teams are working to develop next-gen AI architectures. At Stanford's lab, Chris Ray's team, including architectures like S4, Monarch Mixer, and Hyena, are making strides. The latest (and perhaps most promising) architecture is Mamba, which is more compute-efficient than Transformers. So whether it be Mamba or other models, the AI community looks forward to the emergence of an alternative model. Forbes predicts that in 2024, one or more of these challenger architectures will achieve a technical breakthrough and truly get adopted by AI models.
6. AI Regulation: Judges and Legislators Will Increasingly Influence the AI Industry.
As AI's power grows, questions around AI safety, risks, regulations, and public accountability will come into focus. Hence, many experts believe 2024 may see a slew of AI regulations introduced.
At the A.I. Safety Summit in the UK, 28 countries agreed to cooperate to prevent "catastrophic risks". After marathon negotiations last December, the EU drafted the world's first legislation restricting AI use, including limits on facial recognition and deepfakes, and defined how enterprises can use AI. The final text will be published in early 2024, with the EU's 27 member states hoping to approve it before June's European Parliament elections. We also see early-stage US government participation in President Biden's recent executive order. 2024 will be a year when clear AI governance laws are formulated, happening at both national and global levels – also a test of enterprise agility.
7. Healthcare AI: AI Will Continue to Permeate Healthcare.
"The next big thing in 2024 will be the continued convergence between biotechnology and AI," said Deven Parekh, Managing Director at Insight Partners.
Experts agree that in 2024 and beyond, AI's role in healthcare will expand, enhancing disease prevention, diagnosis, and treatment. For instance, by analyzing medical images, patient records, and clinical data with AI, healthcare providers can make faster and more informed decisions. In the operating room, computer vision can track medical devices and doctor hand gestures to drastically lower the risks of surgical errors.
Beyond these basic applications, we may also see a paradigm shift – AI in healthcare may pivot toward disease prevention. With the help of AI, data, and advanced diagnostics, everyone will understand their biology, tendencies, and interventions for staying healthy more profoundly. The fusion of AI with the field will make healthcare "more accessible, affordable and higher-quality".
8. AR & VR: AI May Herald New Era for Virtual and Augmented Reality.
"With the recent developments in the area of 3D scene reconstruction, such as NeRFs and Gaussian Splatting, the future of augmented reality looks brighter than ever before," said Michał Tadeusiak, Director of AI at deepsense.ai.
Thus far, the fruits of generative AI models have been mostly through chatbots and Discord servers. With the rise of multimodal technology and progress in computer vision, leading AI companies will double down on AI-first wearable devices. In fact, we already see companies like Meta and Microsoft exploring headsets and glasses driven by large models.
Information believes that in 2024 they will emerge into the real world via wearables and robots leveraging AI to recognize objects and showcase multimodal abilities. Such integration will enhance experiences in manufacturing, retail, education, and more – providing immersive education and shopping as well as operational support. Healthcare professionals may even use AR to guide or even perform remote surgeries.
9. Real-Time Computer Vision: CV Will Focus More on Edge Computing.
Edge AI is bringing new possibilities across sectors. According to viso.ai’s prediction, computer vision in 2024 will emphasize edge computing. Processing visual data directly at edge devices capturing the data (e.g. smartphones, drones, and IoT sensors) reduces latency. This enables real-time visual data processing.
In applications, real-time computer vision can enable real-time feedback during remote surgeries, or get leveraged for security, crowd monitoring, and industrial safety. Notably, it enables real-time decision-making in autonomous vehicles. A prime example is Tesla's Full Self-Driving (FSD) beta program, which utilizes deep learning models on edge devices to achieve advanced self-driving capabilities. This innovation highlights edge AI’s ability to perform real-time, complex tasks - laying the groundwork for widespread adoption of self-driving vehicles soon.
10. Data: Enterprises Still Need More, Better Data
Yes, data was, still is, and will continue to be the gold mine for AI development.
In a recent letter, Andrew Ng wrote that "good data for AI systems to function well" is something that won't change over the next decade. Just as humans need good data to make decisions on which marketing tactics to pursue or what our kids should eat, AI needs good data as our algorithms expand, evolve, and improve. There needs to be continued cultivation of data-centric AI practices.
A year ago, Forbes predicted we are running out of data to train large language models. Unfortunately, this prediction has proven true as we rapidly approach the limits of available text training data. OpenAI's "Data Partnership" initiative seeks new training data sources to avoid this impending crisis.
As multimodal models rise swiftly, real-world images, videos, point clouds, and other multimodal data may also eventually encounter text data difficulties. Hence enterprises should value acquiring and maintaining data assets. At BasicAI, we're keenly monitoring this trend, striving to offer training dataset creation services for enterprise AI teams, while continuously upgrading our data annotation platform to meet efficiency, quality, and multimodal demands.
These are the leading AI predictions for 2024. We, at BasicAI, hope 2024 will continue bringing the excitement of discovery and sustained innovation to us all!
If you're exploring the AI frontier, come talk to us about data solutions for your next AI project in the new year.
Read Next
Data Annotation in 2024: Shaping the Future of Computer Vision
Instance Segmentation: Comprehensive Guide for 2024
The Foundation Model: Key Facts and Insights
Computer Vision Unveiled: Navigating its Evolution, Applications, and Future Horizons
Machine Learning for Medical Imaging Analysis: A Comprehensive Overview [with Datasets Map]
