Rob is the CTO at Applause, helping leading brands quickly release high-quality digital experiences.
My previous two Forbes Technology Council articles covered AI because, regardless of whether you are a CTO, medical professional, college professor or musician, chances are you have heard about, and maybe even interacted with, AI. The conversation around AI’s promises and pitfalls will remain front and center as we start to see some buyer’s remorse and value questioning on the negative side, but promising new applications and improved productivity on the positive side.
Where We Are Now: The Difference Between Generative AI And AI
In 2023, we saw that although the world is ready for AI, in terms of mainstream use, we’re not quite there. Generative AI has been the subject of scrutiny, partly because in some situations, it is not providing the solutions the world wants just yet. There are many impressive things generative AI can do, but it has its limitations, which can be problematic depending on the situation.
Let’s consider the tendency to hallucinate (give answers that seem like they could be correct, but turn out to be slightly or even blatantly wrong). While we can think of examples where providing incorrect information may not be that big of a deal, what about in situations like healthcare, legal briefings or engineering equations? Mistakes in these professions can not only cost someone their job but can have deadly consequences.
However, while we’ve seen plenty of examples of generative AI’s shortcomings, next year will likely be more about trying to determine the use cases for where it works best and finding solutions for improving the areas where it doesn’t work so well. To that end, we’ll start to see much more domain-specific generative AI, in which the large language models (LLMs) used to develop these applications are specialized for specific verticals.
We’ll also start seeing more applications using generative AI combined with actions. In the past, we operated with a simple “Alexa, turn the bedroom light on,” but in the future, it could be something like, “Please get me access to Salesforce,” which will trigger a whole series of events, approvals, communications, etc. We will be able to give increasingly more complex instructions and the applications will be able to execute more of them over time.
Today, generative AI is mostly regurgitation of text and images, sometimes in new ways, but bereft of actions. They’re coming, and the combination will start to blur the AI lines.
The Next Phase: Enterprise Adoption, Talent Management And Accountability
One area where generative AI had a big impact was in enterprise organizations, and in 2024, we’re likely to see an even more rapid adoption of the technology. Early adopters now have a solid tech stack foundation and a better understanding of the risks and benefits of deploying LLMs. They will become more comfortable incorporating their proprietary data into LLM-based applications and will have an evolving understanding of the business value AI can drive across different functional areas and use cases. We’ll see users developing small internal use cases into larger-scale deployments that can serve enterprise customers and partners.
However, as AI grows, so will concerns about accountability. From an organizational perspective, enterprises will need more data scientists, prompt engineers and AI compliance officers, plus domain-specific experts to help ensure that organizations have all the enabling policies and procedures in place for deploying applications using LLMs. An LLM’s level of understanding is dependent on the data it’s trained on, so enterprises will need to benchmark AI models and test the LLMs’ performance to evaluate whether it adequately suits their specific purposes.
Balanced Provided By Testing And Validation
As organizations grow their AI use cases, they will need to prioritize quality and not just cost and efficiency. Collecting data and training models with the correct data is key. Using the example above of a legal brief, an organization can input all of its existing legal references into an LLM, but the validation and quality check at the end (to make sure that the outputs are valid) can be really difficult without a domain-specific expert like a lawyer or paralegal. Comprehensive testing to cover a wide range of scenarios is going to grow in importance and complexity.
The world has shown that it’s ready to embrace AI, and while we’ve had some bumps along the way, the possibilities are becoming realities. From facial recognition technology and banking chatbots to household voice assistants, AI is becoming more and more ingrained in our daily experience.
But, as is true with any new technology and all applications, thorough testing is key. Taking precautions to eliminate biased, harmful or incorrect responses is essential to the successful future of AI. Organizations that take a thoughtful approach and stay focused on providing quality experiences for all users will lead the way.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
