Should Businesses Be So Quick to Adopt Generative AI?

It is safe to say we are entering a new era of technology. Generative AI programs, such as DALL-E, Bard and ChatGPT, have dominated headlines, amazing users with their ability to churn out detailed, human-like responses.

ChatGPT, in particular, has soared in popularity in recent months, attempting to solidify itself as the most advanced, publicly available natural-language processing tool. While the technology is not ‘new,’ the chatbot’s ability to provide highly-detailed responses across a variety of subjects seems to have earmarked the beginning of a new chapter in AI development, even passing the famed Turing Test – a test devised by the great mathematician Alan Turing to see if a machine can convincingly mimic human speech and be classed as ‘sentient.’

This test, of course, is somewhat outdated, and ChatGPT is not sentient at all. The most impressive aspect, however, is that it is convincing. And that makes it capable of supporting businesses in all sectors, with tasks that would have previously taken a decent amount of thought. From fraud detection and financial reporting to events planning and email campaigns, ChatGPT’s capabilities look boundless.

However, with this comes significant risk. To be sure, implementing any new system in a business must be conducted with caution, but ChatGPT seems to be an exception, with businesses across the globe keen to integrate the tool into operations ASAP.

There are some obvious upsides to the technology, especially following the latest release of GPT-4. For example, ChatGPT now has the ability to analyze and describe images, an enormously valuable capability for those that are visually impaired or blind. The Be My Eyes app has recently incorporated GPT-4 into a “virtual volunteer” to describe the objects around a user, helping those with low or no vision interpret their surroundings.

Caution in using even for mundane tasks

However, it is not clear how its new capabilities will perform, having only just been released for public testing. It is no secret that OpenAI has emphasized the fact that ChatGPT’s information will not always be infallible. For this crucial reason it is important that the chatbot is thoroughly tested before being rolled out across any business and all testing should be conducted in a sandbox environment, to limit the risk of the technology in your day-to-day operations.

Realistic applications of GPT need to start in areas of high value, high feasibility, and low risk for all stakeholders, including, at a minimum, mundane and less risky administrative work.

A huge oversight by many businesses is that ChatGPT cannot access the internet or external links. Its information is limited to the text it was trained on, which OpenAI has stated only covers information up to 2021. Not only this, but the data used to train the chatbot was written by humans and is therefore susceptible to human biases implicit in society.

Therefore, there is a real risk of spreading misinformation. And it is not just general ‘fake news’ at stake here, there are demos online of ChatGPT’s ability to generate seemingly sound legal agreements. Companies quick to adopt this technology for legal use could risk putting themselves in serious danger if not used with extreme caution.

This comes with the added threat that any business over-reliant on the technology could risk making their operations dependent on ChatGPT’s services. This became evident a week after GPT-4 was released, when ChatGPT went down for over a day, preventing users from accessing the service.

Going a step further, ChatGPT has the ability to work against businesses, and can be used to write convincing phishing messages and create malicious code (even without prior programming knowledge by the prompter). It can also generate working code for adversarial attacks, ransomware attacks, and other exploits when given a simple prompt, arming potential hackers who earlier were limited by their programming knowledge. In fact, 51% of IT professionals predict that we will witness a successful cyberattack with the help of ChatGPT by the end of this year.

Are businesses willing to take responsibility?

Considering these potential risks, should businesses realistically consider testing and implementing the technology?

With a firm understanding of the technology’s own biases and naivety, ChatGPT becomes a tool of immense potential. As long as you are prepared to look and account for its limitations, there is seemingly no end to what generative AI can be used for.

At present, we are seeing the capabilities of generative AI in its infancy, generating highly detailed imagery, copy and audio through human prompts. However, it should also go without saying that this technology will improve over time. ChatGPT Plus is a newly updated system that its maker claims provides more creative responses, is less likely to make up facts and is less biased than its predecessor.

All businesses and sectors, and humanity itself, can benefit from AI, as long as we embrace it with a healthy dose of caution.