Is it ok to use ChatGPT?
By identifying a set of harms, we can give a simple methodology for determining if it’s ok to use ChatGPT in an application. Some rules for the implementation of the applications that are developed are provided. And I can confirm these blogs are 100% human created!
Technologies can break loose from their creator’s control. Mobile phones are a great example. Bell Labs invented them, and for thirty years the telecoms industry saw them as extensions of landlines. Their view was that a mobile phone would let you make and receive phone calls when you were away from a fixed line phone. In the 1990s, telecoms discovered mobile messaging (SMS) and a revenue bonanza turned companies like Orange, Verizon, RIM and Vodafone into technology giants.
Then the iPhone was introduced, and suddenly phones became not phones, but portals to an internet of applications and services. Mobile social media bloomed and mutated into something that no one at Bell Labs would have recognised or countenanced, and the rest is (ongoing) history. The telecoms industry was out of the picture and Apple and Google now define the product and use cases for mobile phones.
This kind of escape can have unexpected and unwanted consequences. Many have argued that some of the new applications of mobile phones are hugely damaging. For example, applications such as Instagram and Snapchat (consumed via ‘phones’) are sometimes blamed for destroying young-women’s self-image, leading to a rise in suicides(1).
A new and powerful technology
ChatGPT is a new and powerful technology that is also escaping the control of its developers.
Until a few months ago, chatbots were narrow and brittle interfaces onto information sources. Applications employing them were expensive to develop and often failed in practice. For example, Facebook launched a service called ‘M’ in August 2015. This was slated to act as an automated personal assistant, but in practice, by 2018, Facebook was only able to achieve a 30% automation rate for the service.
In contrast, ChatGPT is a chatbot that has broad abilities in terms of handling different types of conversation and conversing on different questions. It appears to be very flexible and can be adapted to a variety of tasks cheaply and quickly. Because ChatGPT has become available and the methods used to build it are so well-known, there is now a rush from other technology providers to launch similar technology, and a legion of independent developers is creating applications using it. We have to get used to the idea that generative conversational assistants now exist and all sorts of people are using them.
My new GFT thought leadership paper ‘Using ChatGPT safely’, tries to answer the following questions: So what? What applications are off limits? What should be done to make things safer?
Impressive feats of natural language processing
These tools are underpinned by a generation of machine learning models collectively known as Large Language Models (LLMs). The first of these was the BERT model(2) in 2018 and since then LLMs such as ChatGPT have grown to be 1000 x larger. As LLMs have grown, they have become more and more capable of impressive feats of natural language processing. They are now able to generate long form text, poetry, computer code and interactive conversations.
This is a bit of a shock. There are several reasons why it has happened. Firstly, huge amounts of money have been poured into the effort, allowing a ‘brute force attack’ with vast amounts of cloud compute used to power LLMs. The training methods developed for the first generations of LLM have also turned out to be very inefficient, and very clever people have created much more efficient ones. Finally, and possibly most troubling of all, the life blood of machine learning models is data, and the creators of LLMs have (with some honourable exceptions) demonstrated rapacity and ruthlessness to get it.
Despite what it says when you ask it, ChatGPT appeared at the end of November 2022. Already, just a few months later, many unexpected applications for ChatGPT are appearing. These range from the apparently ethically unproblematic to the obviously malicious. In my paper, I review the application areas that are emerging and outline why these might be considered problematic (or not) and then I will outline what steps can be taken by business people to decide if an application is appropriate and ethical, and what mitigations can be put in place to make fielding something possible.
Discover what’s off limits and how you can make ChatGPT safer in part two of this blog
ChatGPT: Making things safer
It is pretty clear that we are not going to put ChatGPT back in the bottle. The techniques used to create it are well known, and although the amount of compute required seems to be heroic now, in the relatively near future it will be much more widely accessible. Even if compute prices do not shift down radically in the near future, the kind of compute required to create GPT3.5 is already available to many state actors, and a wide range of non-state actors.
Google has announced ‘Bard’ based on its LAMDA technology which is so compelling that one internal engineer became convinced it had a soul and Deepmind has developed a chatbot called ‘Sparrow’ which is ‘claimed by some’ to be technically superior to ChatGPT.
The big dangers are not likely to come from sophisticated super companies like Alphabet. Smaller companies with a ‘move fast and break things’ attitude are likely to be creative and adventurous with their application ideas. But very real harms to very real people are possible with this kind of system, and these can be easily and quickly implemented by small nonexpert teams.
Five top tips to make ChatGPT safer
Even though there are many paths to ‘no’ and only one to ‘yes’, there will still be a lot of applications that get qualified as reasonable. But this will not make them safe. In order to have confidence in a ChatGPT-powered application, it is also suggested that the following steps are implemented.
- There should be no deception about what it is that users are interacting with. You cannot give informed consent if you are not informed. Saleema Amershi et al* have published excellent guidelines for interaction for AI systems. Importantly, these provide structure for considering interaction throughout the lifecycle of a user interaction. The guidelines cover how to make it clear to the user what they are interacting with and how to instruct them about what is expected of them. Amershi’s guidance extends throughout the interaction, managing failure and overtime as the system becomes ‘business as usual’.
- Users should have the option to not interact with the system. A real option – for example an alternative contact channel.
- There should be an impact assessment attached to every application. Put it on the website as you would a robots.txt file, or as you would add a licence to your source code. The Canadian AIA process offers a model for this sort of thing, but some fundamental questions are a good start. Who will it hurt if it works as intended? Who will be hurt if the chatbot goes wrong? Can anyone tell if the chatbot is going wrong, and can they stop it and repair the situation if it is?
- If your system could have an adverse effect on others, then there should be monitoring and logging of what the system is doing and how it is behaving. These should be maintained in such a way as to allow forensic investigation of the behaviour of the system, if required.
- If you are not personally and directly responsible for the system, a clearly documented governance process should be developed and maintained. Part of this should describe how users can call for help, and how they can complain about the system. It should also describe what the processes around addressing user distress and complaints should be.
Potential for great value in many use-cases
In my new GFT thought leadership paper ‘Using ChatGPT safely’, I have laid out what the potential problems are with ChatGPT-based applications and some tactics for avoiding and mitigating them in practice. I hope that our community deepens and develops these approaches in the near future.
With the correct controls and processes, new large language models such as ChatGPT will provide great value in many use-cases, albeit with the essential controls and checks in place, to ensure users and end-users are protected from any misunderstanding.
*Amershi, Saleema. ‘Guidelines for Human-AI Interaction.’ CHI conference on human factors in computing systems. CHI, 2019. 1–13.
