Artificial Intelligent (AI) technologies are reshaping the way we interact with various personal devices and are providing enhanced contextual user experiences, but does this mean we can trust it?
Whether it is a voice-enabled smart speaker that will recite a recipe while you are cooking or play your favorite movie on your television, AI is impacting many aspects of our lives in meaningful ways. OC&C Strategy Consultants estimate that voice shopping is expected to surge to $40 billion within the next five years from $2 billion today, engulfing all aspects of our lives essentially.
This amplification of the role of AI has given rise to a series of questions, but in my opinion, the most critical one is:
how much should we trust such AI systems?
As I mentioned in my previous blog post, Infusing AI to Accelerate Digital Transformation: The (Human + Machine) Collaboration, I feel that AI is increasing the diffusion of ‘Smart’ technologies, amplifying the human-machine relationship and establishing a foundation for the next generation of problem-solving ecosystems. And to realize the full gamut of benefits from AI, we will need to trust it.
Trust is the building block of any relationship, between humans and machines or between machines only. For humans, trust is knowing that the individual who is trusted will do what is expected, creating a feeling of safety. But in this case, most people are unfamiliar with AI resulting in the lack of confidence and acceptance of AI.
Tackling Machine Bias
We are at an inflection point where AI makes decisions that affect many aspects of our lives. You may be comfortable with insight-driven recommendations while searching for a product or service. For example, chatbots that suggest loans based on your creditworthiness, driverless vehicles, AI-powered medical devices and more. AI biases play a critical role in personalizing recommendations to the users, and the AI platforms reference data plays a critical role in the overall ability of the platform to offer personal and useful suggestions to the consumers. While models that are implemented by AI are more global in nature, personalizing these models are done based on micro – models based on personalized data collected from the consumer and Bias neutralization is key to build trust from consumers in using these platforms.
Hence while integrating AI, it is not only critical to instill the capability to explain their decision-making process. Systems that will ‘learn and reason’ via cognitive computing techniques will allow humans to evaluate the system’s rationale. The method of training AI systems and infusing the ability to explain its actions will bring transparency to machine-learning algorithms, eliminate the possibility of unintentional biases, produce AI systems that are highly accurate while increasing algorithmic accountability.
Explainable AI (XAI) In-Focus
Let us look at a near-real example of an autonomous vehicle of ‘Explain-ability’ where it is important that algorithms explain how they make decisions. Autonomous cars can prevent collisions by detecting what is happening around the vehicle. Imagine a situation when an autonomous vehicle senses that it would collide with an old man crossing the road. The car must make a quick decision, whether to crash into the old man or save the passengers in the vehicle. As humans, we would make the choice of saving the old man as well as the passengers in the car, and hence the algorithm should have been programmed to adapt human behavior fit for the new circumstance, and this is where efficient programming of decision making is crucial.
Another area of XAI is rewiring concepts of medicine and healthcare delivery. Introduction of AI in the medical field does not mean automation of manual tasks performed by medical experts and the freeing up of their time, but an increase in productivity and efficiency. As time is the most valuable resource in medical diagnosis, delaying effective diagnosis will, in turn, delay the correct treatment. Being able to detect diseases in its early stages with the help of AI can be very advantageous, but when gone wrong it can lead to many complications.
Here it is essential to identify and bridge the understanding gap from what might be possible to what is going on with one's health. Such rationale will outline the tools and procedures that the clinicians must use to offer the best possible medical treatment/solution. Knowing the algorithms or the logic behind the decision-making process will not only enhance a medical practitioner's comfort and confidence in their approach but will also help patients understand and trust the line of treatment.
The Onset of Trusted AI
As AI-enabled devices have made their way into our everyday lives, changing the way we work, play and relax, it is vital for manufacturers to design devices that would be adopted based on consumers' trust.
Smart homes are now the norm, and with every aspect of a home becoming connected and more intelligent, voice-based interactions and conversation controls are gaining popularity. Also, they are not only getting connected with effective data rapidly - data around usage, frequency, time of usage, etc., but are further offering customers excellent recommendations from a usage standpoint.
Let us take the case of connected bathroom experience while brushing the teeth in front of a mirror. The intelligent toothbrush and smart mirror will analyze and provide suggestions to the user as well as the manufacturers. Consumers can control features of a smart toothbrush based on their brushing patterns so that their tooth shine, the rigor can be adjusted for longevity thereby focusing on both health and cosmetic value, and finally, the acquired data could also help manufactures improvise their design.
A connected brushing experience may seem ultra-modern, but if you take a closer look, it presents personal information of consumers to organizations. Information such as when a consumer is home, details about their everyday routines including meal times, personal hygiene preferences and more. While consumers might enjoy and feel sophisticated to live in a such a modern, connected environment, it does ask them to place a considerable amount of trust with the AI system.
To build trust between humans and machine empathy is the most essential ingredient. Machines will have to be trained to learn, mitigate bias, understand the users, recognize their emotions and provide befitting solutions.
Going forward organizations embracing transparency of AI along with the ability to explain how decisions are being made will gain the confidence of their consumers.