Digital assistants and conversational AIs are a common part of our daily lives. Industries like banking, insurance, telecoms, and retail sectors all use AI to interact with clients, not to mention in emerging areas like autonomous vehicles.
Of course, we have all seen movies where artificial intelligence systems develop a sense of self and begin to devastate the planet. The AI will happily serve our needs one moment, and then send red-eyed cyborgs through time tunnels to end existence as we know it the next. Even though we are aware that these dystopian stories are the stuff of blockbuster films, it is also true to say that these dramas are, in some ways, influencing how we perceive what AI might be able to accomplish in the future.
In this blog, we will get to know
· What is meant by “responsible AI”
· The significance of developing responsible AI
· The guiding principles of Responsible AI
· A Responsible AI for the healthcare sector
What Is Responsible AI?
Responsible AI means designing, developing, and deploying AI solutions with the core principles of empowering employees and businesses. In addition, the AI’s design should equitably impact customers and society, allowing companies to engender trust and scale their AI confidently.
Importance of making responsible AI
As the tech environment evolves from text input requests to voice-enabled AI-powered chatbots, how much thought do users give to the ethics that underpin how companies build their digital assistants?
Consider, for example, the findings from Accenture’s 2022 Tech Vision research, which showed that only 35% of consumers worldwide expressed confidence in how businesses implement AI. And roughly 77% of those surveyed believe that organisations should be held fully accountable for any misuse of AI.
Sundar Pichai, the CEO of Google, echoed this idea when he said, “There is no question in my mind that artificial intelligence needs to be regulated. The question is how best to approach this.” The EU is one organisation looking to answer Pichai’s question with its working framework called ‘Ethics Guidelines For Trustworthy AI.’
While the EU consultation is still a work in progress, the message is clear. As AI use accelerates, legal frameworks will compel developers to build their AI products based on ethical policies that belay users’ and society’s understandable concerns.
Guiding principles of Responsible AI
AI and the machine learning models that support it should be comprehensive, explainable, ethical, efficient, and built on well-documented principles.
Comprehensiveness — an AI has clearly defined testing and governance criteria to prevent machine learning from being hacked.
Explainable — ensuring AIs are programmed so we can describe their purpose, rationale and decision-making processes in a way the average end user can understand.
Ethical — AI initiatives have processes that seek out and eliminate bias in machine learning models to avoid undesirable distortions in outputs or intentions.
Efficient — AI can run continually and respond quickly to changes in the operating environment while minimising environmental impacts.
Responsible AI is vital in a Digital Assistant
AI has the potential to revolutionise how we lead our lives. It will increasingly change how we interact with business and make complex decisions on our behalf as the tech becomes increasingly sophisticated.
In short, we must trust that an AI digital assistant works in our best interests using optimised, unbiased data and in a scrupulously controlled data architecture. Key processes we must create are AI digital assistants that are interpretable, fair, safe and respectful of user privacy.
Responsible AI, therefore, is an aspiration aimed at doing precisely these and should include a focus on the following:
A digital assistant’s functionality should consistently meet user needs. The principles of human-centered design and comprehensive user input in the build stage are crucial to this criterion.
At launch, implementing feedback loops, system metrics, confidence scores, and continuous learning ensure that the assistant continues to deliver against its core purpose.
When we at phamax began with Ariya as a pilot, it was all thanks to the innovators who partnered for trial. We were able to discover how crucial the user journey is. In addition to simplicity, the interface should take care of ease of access.
The criteria for any digital assistant must account for competing compliance frameworks, especially in complex areas like healthcare, where standards differ between geographies. This will inevitably lead to trade-offs.
With the increasing demand for information in digital space, it becomes inevitable to take care of sensitive information. Especially in channels and webpages that are patient-centric. For example, in some countries, users should be able to get information approved by their national regulatory bodies; therefore, a digital assistant’s design must consider these variations.
The process of content approvals and application logic to ensure the right content delivery may be difficult but is necessary when designing the embedded Ariya version.
Data warehouses are varied and businesses prefer their own data management systems. Especially in healthcare sectors where data sensitivity is more, organisations tend to resist any external systems that can breach privacy. Hence, digital assistants should be able to account for these challenges and should be able to connect to any format of client business data sources, including external datasets. A secure architecture is, therefore, an essential prerequisite to data integrity. Furthermore, the source data must be used objectively.
Ariya is designed to leverage existing data sources and can be hosted in client-preferred environments and channels. The recent introduction of Ariya on MS Teams has proved to be a boon as organisations can limit their access to internal as well as access and connect to their own ERP systems
Ensuring AI digital assistant providers inform users of their rights to privacy in data storage and usage, including in improving performance and user experience. Unfortunately, the potential for a poorly designed digital assistant to contravene these rules are legion, so robust privacy protections is paramount. These rights are outlined in the EU’s General Data Protection Regulation, and failure to comply with it can result in severe financial penalties.
For example, what measures must be taken to protect people’s (employees, patients, and stakeholders like KOL/HCPs) privacy, given that ML models might recall or disclose details of the data they have been exposed to? What steps are needed to ensure users have adequate transparency and control of their data?
Fortunately, by applying several strategies in a precise, principled manner, the likelihood that ML models reveal underlying facts can be reduced.
Controlled Access & Administration
The rigorous control of access to a conversational AI across broad user populations is essential. Sales managers, for example, should be able to view all user data, but a sales rep, for example, cannot view data of other representatives or their territory.
Also, interactions and communications in the digital space have gained popularity amongst KOL and HCPs. It is crucial to distribute region-specific information. This process might include implementing a secure login for accessing patient data, such as DocCheck, that restricts access only to registered healthcare practitioners.
Communicating to stakeholders about the digital assistant’s limitations to users. The data models a chatbot uses should broadly reflect the patterns within the data used to train the AI. As a result, tech firms should be able to effectively communicate their data model’s scope and coverage, clarifying both its capabilities and limitations where required This will undoubtedly be a typical “moving feast” as new data enters the systems. Therefore, suitable levels of ongoing review and governance will also ensure a digital assistant performs reliably, maintains its objectivity and meets its core functionality.
A Responsible AI For the Healthcare Sector
As AI technology becomes more prevalent, we’ve learnt that the ethics of responsible AI should always stay front and centre on all things AI. phamax being at the forefront of conversational AI use in the healthcare sector, takes these concerns seriously and has ensured these principles remain a priority as Ariya, our AI-powered digital assistant, is being developed and upgraded.