Predictions Series 2022: Redefining AI in a Human Context
Redefining AI is a very practical task. Artificial intelligence is an adaptive technology. And, as the technology has developed over time, its definition has shifted with its use cases. AI is an overarching term that incorporates machine learning (ML), computer vision, deep learning neural networks, natural language processing, time-series analysis, prediction, recommendation and more. These technologies can grant superpowers to their users. But only if used correctly.
Redefining AI for 2023
While AI can sometimes act in autonomy, more often than in a commercial setting, AI is most productive when it augments existing human capabilities. By interpreting AI in this human context, organizations can maximize the benefit they derive from it.
NLP Insights: Infobip Creates AI-powered Chatbot for Uber
AI and superhuman abilities
Artificial intelligence is not typically fully autonomous, but it can work alongside human judgment in order to circumvent human limitations. The healthcare industry offers clear examples of how this works.
When a patient visits a physician, the potential diagnosis and recommended treatment is limited by the experience of that particular doctor. They can only draw upon their training and experience lots of potentially valuable information is missing.
But using AI to pool all of the collective medical experience around a particular condition can improve the likelihood of treatment success. A doctor is then able to recommend the best intervention based on the entire recorded history of treating the condition. In the recent case of the COVID-19 pandemic where variants and treatment options were evolving on almost a daily basis, access to the most current information could be particularly influential.
But, here’s the catch – AI is not infallible.
An experienced practitioner might spot a particular symptom during diagnosis that was missing from AI training data. So the optimal scenario is providing humans with the insights that AI can generate while ensuring that people remain in charge of the final decision.
A product ethos for introducing AI
Let’s look at more general best practice for rolling out AI models. If this is something that your organization is considering, a product mindset is best.
AI should serve the people within an organization and help them hit specific targets and goals – it’s essential to keep this in mind throughout the lifecycle of an AI solution. Most leaders understand what data is available to them and how they should be able to benefit from it, at least theoretically.
A CTO, for example, might be tempted to jump straight into developing a data architecture with cost and timelines etc. But alongside these technical considerations, there should always be a human track, where the people the technology will serve are interviewed about their needs and use cases. This ensures AI spend can be directly tied to ROI for its users.
Transparent AI serves people best
Often transparent AI is seen as concessionary – that by reducing complexity, you also reduce performance or accuracy. And in some cases, especially in generative AI, deep learning or a black box approach might offer a more “accurate” result. But in a Human-in-the-Loop approach, transparency enables a better, more balanced result through human + machine outcomes. Transparent AI is the only real route to delivering ethical AI, which in turn is the only way that the technology can be sustainable.
In an increasingly ethically-conscious world where retaining consumer trust and achieving sustainability is central to an organization’s survival (as well as that of humans, in general) there’s a real economic value to AI transparency alongside the ethical ones in preventing bias. In achieving these outcomes, we improve both economic and social impacts.
AI ML Insights: Got It AI Announces AutoFlows, a Breakthrough Autonomous…
Emerging technologies enable model explanation, (and more critically data explanation), which makes it possible to achieve transparency within AI. Additionally, these technologies make outcomes explainable, which enables prescriptive AI.
For example, when building a churn model to predict whether a customer will stay subscribed to a service or not, it’s possible to establish a score from 1-100 of how likely it is that a particular customer will leave. But it’s much more valuable to also understand how that score was derived.
If you can understand the specific detractors and promoters that led to an individual’s score, you may influence their decision. For example, if price point is a factor for an individual customer, you can offer that customer a discount, but not offer it to others who have less price sensitivity.
Artificial intelligence is a powerful tool for modern, forward-thinking organizations: ethically, sustainably and economically. But only if its implementation is based on maximizing the potential of performance within human systems. Without human plus machine approach enabling a product outcome, it will be difficult for these technologies to make a measurable and sustained impact.