ChatGPT, Bard, DALL-E 2, Midjourney – there’s a plethora of generative AI tools with new and exciting use cases out there.
Generative AI has the potential to be one of the most disruptive technologies we’ve seen in some time; however, there are still reasons to be wary in terms of relying too heavily on it at this stage in its evolution.
As with any new technology, there were some wobbles at the start. But these tools are arguably starting to become reliable enough to use commercially, in a professional setting. There are countless ways organisations are now using it to successfully automate or accelerate work and free up resources, allowing them to grow.
Even if some services aren’t production ready yet, we will need to prepare for it, as our competitors are already adopting them.
Content creation
Generative AI can provide a huge productivity boost for businesses that rely heavily on online content or long-form documents.
Consider how much time conveyancing solicitors and lawyers spend drafting what are often standard contracts. They’ll be able to train an AI model on their documents and have it develop new ones based on internal guidelines.
Even professionals who don’t write contracts or other lengthy documents could benefit from this technology. Think how marketing departments could use it to build SEO keywords into materials, or accountants asking AI to create a spreadsheet showing specific expenses.
Generative AI tools use machine learning algorithms so, with a bit of fine-tuning, you can even get the content close to your house style.
But we haven’t yet reached the point where generative AI tools can completely take over content creation.
Of course, any document that has an element of risk will still need human involvement to be made client-ready – but there are added limitations to the technology even beyond that. You’ll need to fact-check and provide your own references, especially since most AI tools aren’t trained with the most up-to-date information.
Still, generative AI can get you about 80% of the way there from a blank page, which will be a game-changer for businesses across all sectors. It can also help you review your own writing for consistency and structure.
Code Completion
People got very excited when GitHub Copilot launched. Since then it has come under some scrutiny when it comes to things like open-source licensing and code privacy. Some treat it as a fancy version of autocomplete, and some see it as an existential threat.
Generative AI has huge potential in “code completion” which I feel is underselling its uses; it excels at finding bugs, refactoring, and documenting old code. Products like GitHub Copilot, Amazon CodeWhisperer, and TabNine are also brilliant at writing test cases, unit tests and translating code between languages.
To be clear, AI is not a replacement for engineers; it’s an accelerator, allowing them to spend their time-solving difficult problems rather than writing boilerplate code. The potential to reduce things like new developer onboarding using this tooling is very interesting.
Chatbots, CRM, and search engines
Automation allows technology to take the lead where possible, freeing staff to focus on the aspects of work that require a genuine human touch.
Ironically, one of the most obvious use cases for generative AI has been with helpdesks – something which even just a few years ago we would have written off as needing that human touch.
The truth is that the use of AI tools in this space falls somewhere in the middle; I’d caution against using a 100% chatbot-run helpdesk, at least for now. We just haven’t quite reached the point where AI can consistently provide the best (or even accurate) information or answers.
However, generative AI can prepare automated responses based on ongoing conversations, so staff only need to vet a response before hitting ‘send’. It would mean less time spent typing or looking up policies, and they can actively participate in multiple chats simultaneously. This isn’t just limited to chatbots, the same is true for other lines of communication such as email and social media messages/comments. In low-risk sectors such as retail automated helpdesks could be used to great effects, such as to suggest clothes or makeup based on measurements or complexion.
It can also be used for CRM augmentation, to automate parts of the workflow, provide customer segmentation, sentiment analysis and hyper-personalisation of communications.
However, industries that have an element of risk inherent in their work such as financial services, healthcare, or legal advice should limit the use of chatbots to signpost users towards information rather than specifically giving advice. Steerability in large language models such as GPT is a key feature here, ensuring that the model “stays in its lane”.
Using it for internal searches on a company’s intranet, especially as generative AI welcomes soft searches, is also a realistic option, recognising someone might be searching for a policy despite typing “process”, for example. It also does a great job of summarising documents for the ‘tl;dr crew’.
Barriers to adoption
Despite the low barrier to entry, there are several reasons businesses shouldn’t be too quick to widely adopt generative AI beyond what I mentioned above.
One of the biggest issues with AI, in general, is bias: A model is only as good as the data fed into it, and generative AI’s lack of explainability could mean users won’t recognise when something is biased.
There are other considerations with using language AI models in a professional capacity, such as hallucination – which is when a Generative AI tool gives a response that seems correct but isn’t.
Another is that it’s not ‘deterministic’: I can ask the same question to the same AI tool at different times of day and get different answers – this raises the question of reliability and how you automate any testing of Generative AI-enabled systems.
Some of this can be overcome by “adversarial AI” – using one AI tool to test another AI’s response and confirm whether it sounds reasonable and attack it to see if it breaks its own rules.
A future iteration of Chat GPT will include fact-checking and referencing – but this requires unbiased, correct data in the first place.
Current use
Despite all these cautions, however, there are already some current, highly successful and effective uses for generative AI.
We’ve been training a chatbot for a client in the personal finance space, to talk about specific topics such as broadband options. However, if a user asks the chatbot about topics outside its remit or training, it refuses to answer. This helps increase reliability and improves the customer experience.
Developers across all sectors are already using AI to develop or clean up code, as they are very steerable and can even explain the purpose of undocumented code and translate it into different languages.
Generative AI is also making great leaps outside of the text, especially in graphics where its ability is developing at a terrifyingly rapid pace – we are sometimes seeing huge leaps and bounds being made in days, rather than weeks or months.
This new era of generative 3D worlds, with its great uniquity and beauty, is enormously interesting – but there’s still a lot of development work needing to be done before it will affect traditional business practices, in some sectors.
Soon enough, if you can describe what kind of world you want to see, technology will create it for you.
