AI has been radically changing the dynamics of many industries in recent years, resulting in a quiet revolution. Companies that invested strategically in AI experienced a greater success rate when compared to companies that used siloed proof of concept. This is reflected in an IDC report for India: "AI spending will exceed the half-trillion-mark next year. As we enter 2023, we anticipate companies moving up the AI maturity ladder by integrating AI into the core business processes, workflows, and customers' journeys."

As AI moves to optimise core processes, the focus on outcomes will increase.

As artificial intelligence (AI) progresses out of the lab and into broader business use, more focus will be placed in the coming year on demonstrating its value. Organisations will start to emphasise measuring AI's success within their operations to quantify the value it adds. In addition, they will need to be fully transparent about their use of AI and how it tangibly links to outcomes moving forward.

The most recent Deloitte global State of AI study states that 79% of survey respondents have deployed AI in at least three applications (up from 62%), with cost reduction as the most frequently mentioned outcome (78%). 94% agree that AI is critical to success in the next five years. However, proving the business value will remain the biggest challenge for starting AI projects and a key one for scaling these out. The biggest challenge will come from scaling and integrating AI into daily operations and workflows.

Key to this success will be the extent to which organisations begin to use AI and automation. We'll begin to see more businesses using this technology like they would an autonomous car in a wilderness - to allow them to navigate rocky, unpredictable terrain and enable them to make real-time decisions in microseconds that consider not only likely future obstacles but also all kinds of real-world constraints, rules, policies and tradeoffs. The result of this is that we'll see more businesses than ever before moving towards the idea of the autonomous enterprise as a means of increasing efficiency, optimising processes, and helping them to do more with less and solve problems more quickly.

Older, less 'sexy' AI will make a comeback.

In a chaotic environment – post-Covid, amidst the war in Ukraine and the energy crisis – businesses will try to tie back AI applications and initiatives to macro top-level strategic goals such as getting closer to customers, improving productivity and effectiveness, and becoming more agile.

At the micro-level, AI-driven automated decisions will be inserted into processes and interactions, optimising multiple goals. This will allow AI to have strategic choices embedded within them, guiding machine learning and producing the right outcomes. These goals include corporate-level goals and empathic objectives of other stakeholders such as clients, partners and employees' personal goals.

To do this, they'll have to look back to some of the older, less 'sexy' elements of classical AI, such as symbolic reasoning and business rules, and combine them with newer, more modern machine learning technology to provide the best possible outcomes. Good 'old fashioned' AI tech, such as business rules, will be required to model real-world constraints and corporate strategies and policies, and all of this will need to operate safely at an enterprise scale.

Compliance and trust will be key in the new age of AI regulation

Responsible use of AI should not be a rat race because no one would win in an increasingly connected global market. Still, the reality is that whoever is the first to move beyond soft ethics and self-regulation into hard AI law will set the standard.

In 2023, the challenge for organisations will be to operationalise compliance requirements and align them with their good intentions. Inventories of AI-based decisions, systems and products will have to be created, with more scrutiny for high-risk ones. Stakeholder involvement will be broadened to all parties involved, not just the AI techies and data scientists. AI systems will need to be monitored continuously, not just when these are designed, for performance, robustness, and fairness. Employees, partners and customers who are more end users of AI-powered decisions and systems will require transparency and explanations of individual decisions that relate to their mortgage application, job offer or insurance claim at a level that is understandable for them. This also means that intelligence within applications and processes will be centralised to allow central control.

Generative AI, a solution looking for a problem

In 2022 there has been an evolutionary explosion of large AI models. These are termed 'foundational' models, and many generative AI models generate content like text, source code, images, video and more. 

So, 2023 will be the testing ground for these models from Google, DeepMind, Meta and the BigScience consortium. These multiple language models are much larger than GPT-3 and downloadable for research or, in the case of Bloom, even completely open access. This has triggered a rash of demos of text-to-video, speech-to-text and translation apps. 

Generating a video of a dog wearing a superhero cape flying through the sky isn't necessarily a high-value enterprise AI use case. However, there are some immediately usable sweet spot areas, such as speech-to-text in marketing, service and operations, or programming and coding support, and in 2023 we can expect more generative AI start-ups to come to the scene to monetise particular use cases by building on top of generic, foundational models. Building and executing the core models are very expensive, and we will also see considerable funding going into companies that provide access to these generic models as a service.

That said, the big players are keeping their cards to their chests. They are claiming they want to research ai safety first or that these are foundational technologies for the long and winding road towards more general artificial intelligence. But there is also an element of wanting to find and commercialise the proper use cases before the others do. 

Photo by DeepMind on Unsplash