What 2023 Holds for Democratization & Adoption of AI?

As the year 2023 dawned on us and the businesses are staring at an economic whirlwind, AI would be more prominent in CXO agenda than ever before. Besides making accurate decision, AI adoption at every echelon of the business will bring much needed agility in decision making giving precious time advantage. The question is how the lengthy AI implementation could be fastened keeping the pace with the growing expectation of AI usage and it’s ROI. There lies the bigger question of AI Automation it’s adoption and democratization.

A recent study by McKinsey revealed that deploying artificial neural networks could account for as much as $5.8?trillion in annual value, or 40?percent of the value created by all analytics techniques. Of course, this have attracted a plethora of vendors making fiercely competitive AI market. However, in the face of scare and costly resources, as CXOs revisit their AI strategies, the winner will be those vendors who can perfectly balance the 3 pillars of AI triad: Accuracy, Cost, and Time.

In my discussion across industries and maturity levels, the following are likely to be the moot points that will drive adoption and democratization of AI resulting into realization of it’s true potential.

Self-service AI: To draw an analogy or parallel with the www revolution, even 10 years before building website was an expert job. Today one can build one’s own website using various tools such as WordPress, Wix, Squarespace, GoDaddy without writing a line of code! Same is true for reporting or visualization technology. Today it’s almost given that most of the business decision makers without any background in BI will make effective reporting tool with cool visualization, again with the help of self-service tools.

Maturity cycle of AI and data science couldn’t be any different. Truly speaking, this is indeed ‘walk-the-talk’ moment for AI vendors. In other words, they need to ‘Build AI to Do AI’. The space of ‘Driverless AI’ is has already been taken up by AutoML offerings. As per this recent research in November, 2021, the future of Low-Code / No-Code AI is going to grow almost thrice in the next 4 years!

https://userguiding.com/blog/no-code-low-code-statistics/

From my personal experience in setting up big captive data science team and deploy data science solution across the globe, 3 factors will drive this exponential trend.

1.??????ROI pressure to faster adoption – Data Science projects are expected to deliver in weeks rather than months.

2.??????Desire by business (specially entry to mid management) to be part of the AI revolution – it’s evident that more than AI itself, a colleague who knows AI is likely to keep you out of job

3.??????Cost of hiring and maintaining a large data science team – It’s extremely difficult to innovate a business cases for the long term to justify onboarding of sizeable captive data science team. Hiring, maintaining, and retaining talent is out of reach for many industries and organizations.

Of-course, smarter CIOs know once the business reaps the benefit of heavy-lifting by Self-service AI, the appetite for nuanced and finetuned model will grow. Good part is, by then the business adoption battle is own and from that point onwards AI investment will be self-sponsored from the benefit of previous projects.

Glass Box AI:

Interpretability of the AI models and AI adoption goes hand in hand. From my personal experience, a ‘good AI model’ is ‘not good enough’! Leading deployment and AI adoption across different business functions and regions, I can say with fair amount of confidence that AI adoption battle is half won if accompanied with business explainability. In fact, often businesses leaders are ready to sacrifice marginal accuracy for sake of interpretability. As a data science leader, the easiest way to get business practitioners in your side is to provide glass-box transparency of the AI models which is most of the time is non negotiable condition for adoption and automation! For example, I’ve closely observed how deploying an AI driven sales forecast solution optimized only for accuracy hit the wall, as the demand planners was struggling to explain the forecast to wider stakeholders and pin point reasons behind change in numbers from one model run to the next.

Of course, interpretability is the hottest area of AI research today. It is extremely heartening to see that AI vendors are making conscious effort to ensure AI results are compliant with transparent, fair, and ethical. Though ‘AI interpretability’ doesn’t aways mean ‘business explainability’ recent innovation resulting into packages such as Shapley and Lime is good development in the right direction.

Automation Driven AI:

Automation is not new. Rule based automation is there possibly from the beginning of industrial age. What is new is, AI has the unique ability to automate complex business decision making process. One of the biggest factors to drive AI adoption will be focusing on the principle of ‘management by exception’. This essentially means that AI does the heavy lifting allowing the decision makers only to manage the outlier cases which are either of high importance or unique. Low-touch or even No-touch not only saves precious time of practitioners, it also keeps it sanitized from the subjective biases.

Express Data Preparation:

As the golden 80:20 rule in data science, having self-service AI itself will not squeeze the notoriously long AI development / deployment cycle. Automating the data preparation will be the key frontier in the coming years. Unifying data from disparate sources, validating, enriching with overarching compliance to privacy and ethics is the elephant in the room. Good news is progress has been made in the filed of Automated Data Preparation (ADP) which can handle the data preparation tasks, analyze data and identifying fixes, filtering out the fields that are problematic or not likely to be useful, deriving new attributes when appropriate; thereby improving performance of the AI models. With newer use cases and varied data format (such as Graph, Elastisearch) being innovated, I would eagerly watch out this is field for fresh development in the recent future.

** Please note opinion and facts stated here is entirely in private capacity and doesn't represent any organization