AI predictions to watch for 2024

Guest post by Adonis Celestine, Senior Director, Automation Practice Leader, Applause

2024 – The year generative AI becomes accountable

Undoubtedly, 2023 was a rollercoaster of a year for generative AI services, and there were some hitches along the way. One notable problem has been the data privacy, compliance and copyright infringement issues that call into question the data sets being used to train the algorithms of LLMs like ChatGPT and Bard. As a result, some companies have worried about the legal implications of integrating the technology into their solutions. Tighter processes and regulations were called for to ensure artists, customers and others had given valid consent regarding data.

However, concerns about adopting generative AI may be lesser next year as AI companies take full accountability for the data used by their LLMs. Microsoft recently reassured users that it will take responsibility for content Microsoft Copilot – its AI assistant – will generate by extending the company’s IP indemnity. Many other providers will follow suit in 2024 with similar commitments. This will ensure that liability isn’t placed on the end-user, but with the provider – a move that will drive greater adoption, especially among larger enterprises that will look to reduce the risk of liabilities.

2024 – The year brands begin benchmarking generative AI

Testing the IQ of an AI will become an essential part of the adoption process for brands planning to roll out generative AI (e.g., Chat-GPT, Llama 2) powered applications and services. The problem is that generative AI is still in its infancy and the services are very generic. Customer-facing brands looking to adopt generative AI tools will need to benchmark the LLMs they’re based on to assess their suitability for specific use cases like immersive shopping.

An LLM’s level of understanding is dependent on the data it’s trained on, which is why its important brands assess the AI models to make sure they’re applicable to their needs. Benchmarking allows you to test the LLMs performance to see if it can suit a specific purpose. However, it’s not as clear cut as benchmarking one service against the other. For example, Chat-GPT already has several different iterations. So, if your needs are quite basic then an earlier version might be the better and cheaper option.

A thorough benchmarking study will be required for more sophisticated requirements. The only technology that can keep generative AI in check and regulated is generative AI itself. Numerous gen AI-based metrics have evolved to measure LLM against criteria like accuracy, precision, ethics, etc. Augmented with human evaluation, these metrics and benchmarking will keep AI safe and explainable.

2024 – The year of the Democratic AI

Gartner recently identified “Democratized Generative AI” as one of its 10 top strategic technology trends that organizations need to explore in 2024. By 2026, over 80% of enterprises will have used gen AI APIs and models and/or deployed gen AI-enabled applications in production environments, up from less than 5% in early 2023. This will lead to more Democratic AI, essentially making artificial intelligence technology more accessible and available to a diverse range of users. It involves removing barriers such as high costs, technical expertise, and limited access to resources.

Next year, there will be more conversations about how “Democratic AI” can be achieved through the development of user-friendly tools, open-source platforms, and simplified interfaces. The conversation around the need for transparency, inclusivity, and accountability in AI systems took off in 2023, and we’ll see that continue as organizations grapple with who their key stakeholders are and the social, ethical, and legal implications of AI. We’ll also see continued conversation around AI governance, ethical frameworks, and regulations to mitigate the potential biases and discrimination that would keep AI benefits from being distributed equitably.

2024 – The year AI gains its edge

Edge computing is now an established technology with estimated global spend to reach over $300bn by 2026. Generative AI will help to spur that growth over the next 12 months as it contributes to the number of new edge computing and IoT use cases. Generative AI will also add speed of thought and decision making to the compute power already located at the edge, powering mission critical applications that need immediate responses to requests. One of the specific use cases for this new generation of ‘Edge AI’ is autonomous self-driving features in cars, such as slowing down on the approach to traffic lights and stop signs. Edge AI will be active in those locations making decisions in real-time, based on what data it’s receiving in live situations.

Edge AI is now possible because of knowledge distillation – training smaller models with less data which mimics the behaviour of larger models. This removes the need to access the entire data set, ensuring lower latency and the need for huge computation power which costs millions of dollars. Much of the compute power is stored on the device itself so half the time the Edge AI can make decisions without the need for a network connection.

Updates can be delivered over the air when connections are live or when upgrades are due. Expect to see Edge AI use cases appearing throughout 2024, predominantly in healthcare and automotive, facilitating faster and more accurate decision making.

Adonis Celestine, Senior Director, Automation Practice Leader, Applause

See more breaking stories here.