- Virtual assistants using natural language processing (NLP) are here to stay as they become more efficient and cost saving for companies that deploy them.
- However, the technology has potential downsides, including miscommunication and furthering inequity.
- Like any form of artificial intelligence (AI), NLP requires thoughtful governance to ensure it’s used for good, and a framework must be developed to optimize the human-machine interaction.
In today’s world, we’re often interacting with virtual assistants, either by speaking to them or by typing. Think about all the people who have Amazon’s Alexa-enabled devices in their homes and are asking these devices to play music and tell jokes. Amazon sold over 100 million Alexa devices in 2018 alone and that year Alexa told over 100 million jokes.
Have you read?
Alexa is a chatbot, a form of AI that interacts with customers via conversation. That’s because NLP enables it to understand humans’ messages and often respond appropriately. In the case of a consumer seeking an answer to a product question, being able to type the question and have a bot immediately return an answer saves time that may have been spent on the phone, waiting for a human representative to respond. In turn, the product’s manufacturer doesn’t have to employ an extra human to respond to routine questions, which saves the company money.
The potential for such virtual assistants is enormous and many benefits are already being realized. However, all the kinks have not yet been ironed out. While building a chatbot is relatively easy, the conversation piece is often harder to get right.
Chatbots: The good
As for applications for chatbots using NLP, the sky is the limit in industries as diverse as healthcare, education, retail, tourism and others. With many people trying to educate their children via Zoom, chatbots can deliver AI enabled education across the world. Some hair salons have been employing chatbots to schedule appointments and they are also being used for scheduling things like airport shuttles and rental cars.
Healthcare presents perhaps one of the biggest opportunities for virtual assistants. Automated text reminders of appointments have resulted in reduced no show rates in the U.S. And in rural parts of the world, chatbots are helping to connect patients to clinicians via digital consultations.
For example, in Rwanda where there are only one doctor and six healthcare workers per 10,000 people, healthbots are helping reduce the heavy demand on health center staff. Instead of standing in line to see an in person provider, patients can access consultations with doctors or nurses over the phone from anywhere in the country. They can receive a text message code for a prescription or a lab test.
In the future, these healthbots will become triage tools that will use AI localized for Rwandan language and epidemiology to allow even more patients to be served. For patients that require a physical consultation, the triage tool will prioritize those needing the most urgent care. The triage tool can also share patient information with clinicians. This makes it easier for them to quickly access what they need to treat the patient.
Through such virtual assistants, more effective use of scarce health resources will be possible. In return, the quality of care will be improved and healthcare workers will be kept in the loop.
Chatbots: The bad
AI is continuously learning but its algorithms are designed by humans who have biases. One of the pitfalls in using AI powered chatbots is the lack of diversity among creators that can lead to biased responses. Often heavily accented users are misunderstood by bots and that can have implications for patients as well as anyone seeking correct information. Poor guidance, incorrect diagnoses and failure to access timely interventions can result in serious consequences. The challenge is to attract more diverse programmers and recognize specific instances of inequity in communications.
Information privacy is also a serious consideration, as is the ability of users to distinguish whether they’re speaking to a bot or to a human. NLP powered virtual assistants are becoming increasingly sophisticated and sound more “natural” all the time. It’s understandable that patients or other users should wonder if they are talking to a human or a bot – or getting medical advice from a physician or a bot.
To address this issue, Stanford University proposes that artificial agents should be required to produce, on demand, unambiguous identification that they are bots. In addition, the proposal calls for including information about the virtual assistant's history of ownership and usage. This information could potentially address tracking concerns and the question of who’s responsible for outcomes.
Implementation of such proposals is bound to lead to further questions, but a comprehensive debate is needed to establish boundaries and keep people and their sensitive information safe. A Deloitte AI Institute and Chamber Technology Engagement Center study details how industry and government can collaborate to help ensure that AI’s benefits are fully realized and concerns over its risks do not dampen innovation.
Chatbots and Virtual assistants: The forever more
The virtual assistant genie will not be going back into the bottle; NLP powered chatbots are here to stay and their capabilities will only grow. The situation presents an opportunity to unite stakeholders from multiple areas such as developers of chatbots and platforms, the medical community, academia, regulators as well as society at large to design frameworks on how to govern them.
The World Economic Forum has assembled global experts from the healthcare, government and private sectors to develop Chatbots RESET, a framework for regulating the responsible use of virtual assistants in healthcare. This framework is tackling legal, privacy and security issues alongside the need for equity and accountability. Its co creators are establishing ethical principles and operational guidelines - which can be adapted for other industries.
In the near future, with effective governance, we’ll see Zoom calls where people are speaking in different languages being translated in real time, with each participant hearing the audio in his or her own language. In meetings where a scribe is needed, NLP will enable the virtual assistant to listen, synthesize and extrapolate the main points. And when a customer calls with a complaint about a product, the chatbot will be able to tell the precise point when that customer should be transferred to a human representative.
The efficiency of chatbots is undeniable as is the cost savings for companies. But beyond that, these virtual assistants are paving the way for workers who would normally be responding to routine questions to be redeployed into more meaningful, skilled and higher-paying work. Companies are apt to see attrition rates drop as a result. Afterall, who likes to be asked the same question 100 times a week?