Our CTO Aaron Perrott sat down with our CFO Kathryn Milliner to discuss the role of AI in the workplace and it’s impact. The blog was produced in-conjunction with We Are PoWEr.
Kathryn Milliner
Let’s start with what your view is on the opportunities AI is creating?
Aaron Perrott
AI is removing a lot of what we would consider toil or jobs that we don’t particularly want to do.
For example, using it as a starting point to write documents for handover/training. Within an organisation document creation quite often falls to the back of our queue because we’re always conscious that we have to do our day to day tasks which take precedence over the documentation.
By using AI to document why we’ve done it, and allowing other people within the team to understand our thoughts, we can improve our training capability and increase the skills and inclusion of the team, particularly with remote working.
Because AI can do alot of the heavy lifting by writing either the framework of the document or the structure of that document. It means that we can save time producing the documentation.
Aside from documentation, any tasks that are repetitive or laborious is where AI has a role to play. Generating emails, marketing material, summarising info, images etc are all areas where AI can save time. Allowing us to focus on where we can add more additional value, where we can use our human capability.
Kathryn Milliner
How does AI facilitate supporting where you’ve got diverse needs that may have had barriers in the past? So for example, a traditional technical document could be incredibly difficult for someone in the business to understand, where can AI help here?
Aaron Perrott
Leading on from the example above, AI can re-write a document from a different viewpoint. This means we can engage a wider audience as AI can help explain terms which might be specific to IT or Finance. It can put a different lens on it. For example it can take a IT heavy document and phrase for a non IT community.
In order to understand how it does this, it’s important to understand a bit more about how AI works.
AI can behave as many different actors. You tell it a role that you want it to be or how you want it to behave. We refer to this in AI as prompt engineering, but in essence you tell AI two different things :
A user prompt is a specific instruction or question provided by the user to the AI system. For example:
“Generate a summary of the latest sales report.” “What is the weather forecast for tomorrow in Paris?”
A system prompt serves as the foundational instructions that dictate the AI’s behavior. For example:
“You are an assistant that helps users with their queries. Always provide accurate and concise information.”
“Your role is to generate creative content based on user inputs. Maintain a friendly and engaging tone.”
The latter allows it to behave as different actors which is how it can provide a different lens on documentation or respond in a different personalised way. You can define the type of actor you want the AI to be to give the right tone or perspective to your data.
You do need to be careful with system prompts. You may need to work on them to get the AI to behave in the right way. The more information you give it, the more it will ‘act’ in the way you want. Many off the shelf AI’s come with a set of ‘actors’ you can already use.
Kathryn Milliner
Do I need to be technical to do this? How does it work?
Aaron Perrott
No you don’t need to be technical. The level of technical capability needed has reduced massively over recent years, you don’t need to be a prompt engineer anymore. You can just tell it how you want it to behave in human language.
If you can imagine you had a child that was a blank canvas and you just said, I want you to behave in this way today. It will just do it. Unlike a human though it won’t do something unexpected, it will behave and act how it’s been told to the best of it’s ability.
Having said that there is something within AI called hallucinations, and sometimes AI can get things wrong and there’s a reason why it does that which I’ll come onto, but it’s not going to do something that you don’t expect.
Kathryn Milliner
Is that a challenge with AI, in the fact that there is a limitation that it’s only going to be as good as the design/data it has been given?
Aaron Perrott
Yes, definitely. You have to use the right AI for the right purpose and you have to be conscious of how that model was designed to behave and also what data it was given to learn from.
Every AI has to learn, before it can respond back. Within AI we often talk about large language models. Simply it’s a store of language. If you took all of the uses of language in the world and created a model of it that would be the basis of a large language model (LLM). When we use the term language we aren’t just referring to text either. We mean text, images, or even physical language, for example driving a car.
Each large language model can only be based upon the data which it’s been fed. For example an AI capable of self driving a car is probably not going to give you a good recipe for a burger because it’s not been trained to do that and it’s not been designed to do that.
So you need to make sure you’re using the right model for the right thing. There are also many generic models.
For example copilot is one people are familiar with in the work place. It’s a generic model which can be trained purely on your work data, and it will answer most questions within the workplace, and that’s what it has been designed to do, working with the Office 365 tools.
Equally with a public AI such as Gemini (Google), you’re going to get a very generic answer to most questions because it’s been designed and built on generic public data, but you wouldn’t trust Google to drive a car. You’ve got different AI models for specific cases.
You can use an AI which operates within your ‘tenant’. This is quite an IT term, but it just means data within your organisation. Rather than responding with public information it only responds with information available within your company.
Equally you also need to make sure that you’re not feeding company information into a public AI and that’s a real challenge for some organisations.
Kathryn Milliner
What are some of the human challenges with AI?
Aaron Perrott
One of the challenges is making sure there is still a career path with opportunities for staff to learn, but whilst AI is performing some of the repetitive tasks and/or writing the documentation it shouldn’t completely replace sitting down with somebody and teaching them.
We still need that human interaction and we need to be mindful that if AI is removing some of what we would call first line or that initial junior roles, then you might not be getting the right experience to move up in the organisation. This could potentially create a skills gap longer term as people aren’t getting the necessary experience at a junior level to have the knowledge required.
Kathryn Milliner
The future of human skills isn’t going to be necessarily disappear. They will become just as important as they are today. In a world where we’ve got AI being used, there’s a dual learning part which is understanding the topic itself, but also understanding the limitations and opportunities of AI and how they work alongside.
Aaron Perrott
Exactly, the big thing that humans can bring is a level of imagination alongside understanding the person that they are dealing with. If we go back to the point of AI being an ‘actor’, this is a key consideration.
When we have an interaction with somebody we change the way we respond as a matter of course. For example the other day I was working with a colleague who was 30 years younger than me. We were talking about work concepts and Christmas. However because I knew there was a 30 year age gap, I was making sure my responses were relevant (where I could!) and using examples, and/or asking questions which made sense.
If I was having the same conversation with someone I already knew, or someone my age I might use different responses. As humans, we will change the way we behave based on who we are talking to naturally. We don’t need to be told to be a different ‘actor’.
We also understand naturally about boundaries and/or how to react. If we speak to 10 different people in a room, we naturally do this, where as with an AI you’d need to tell it be a slightly different ‘actor’, to change it’s responses and/or to make the relevant. For example if someone had children, you might ask what they got them for Christmas where as if someone didn’t, you wouldn’t ask the same question, you would ask a relevant one and AI struggles with this.
This skill is going to become more important. If you assume AI can give you the data you need, it’s your interpretation of the data which becomes critical. Alongside the skill on knowing the ‘right’ answer for the situation/scenario. You still need to have that skill to say, is that the right answer for the person that I’m talking to? Is that going to be the right information delivered in the right style?
Also being able to take multiple data inputs and make a decision off the back of them. Again, that’s a very human skill and that’s going to become more important with AI because that’s going to be the differentiator.
I do think we’ll see a shift in the way we educate as well, as it’s going to become more important to learn how to interpret data, not remember it. We are very exam based but actually finding or remembering data isn’t an issue with AI. The interpretation and knowing what is right and what is wrong is far more important.
Kathryn Milliner
Where are AI’s strengths at the moment in the workplace?
Aaron Perrott
I still think AI is much stronger at interpreting language and processing language than it is with physical things. Back to the driving the car example or monitoring an alarm system or looking at human responses to a camera. It can do all of that now, but its strength is still the interpretation of language, because that’s where it started.
It’s much easier to do and there’s a huge wealth of material that it can draw upon which is why that has filtered into most workplaces now. It’s also the most cost effective way to use AI.
Another example particularly in the IT sector is chatbots and chat responses. This is because it’s a consistent level of question and response. Which can be backed up with easily maintained data with no little room for interpretation.
The more complex the use case, the more expensive it is to use, train and maintain.
Kathryn Milliner
In terms of the future of jobs, based on what you’ve just described, what do you think it means for the roles we have today and the transition?
Aaron Perrott
I think it is going to put a much bigger onus on being able to bring something different to an organisation. As mentioned above, I do have a have concern around junior roles. Some areas of industry will be more impacted than others. For example in the insurance and legal space, the concept of reading legal documents or insurance documents and interpreting them, can be done by AI and I think it is going to replace those roles.
One advantage AI has, is that it’s ‘always on’ and it doesn’t fatigue, so it can complete tasks to the same level of consistency again and again. In that type of task the AI is better than a human. It acts like a machine.
I do however believe that AI will also generate jobs within the sector itself and there’s a huge amount of opportunity. Across training/building and improving AI models.
To give an example within the psychology sector. Alot of work is being done at the moment in order to have AI perform initial psychological analysis in sectors you wouldn’t necessarily think about where you can take initial case load away from highly skilled people to free up time for the critical cases in an over stretched sector.
By using AI to read transcripts but also (where consent is given) having AI watch the interviews and come up with an initial assessment in counselling and therapeutic sessions. Whilst initially you may feel this is a step further, one advantage AI has is the lack of bias towards a participant, it’s only assessing based on criteria. For example using it to pick up certain clues a more seasoned professional might spot when an interview is done by a less qualified member of staff.
It’s an example where there’s a huge opportunity in that space. People with psychology backgrounds, learning prompt engineering. Feeding and training the AI to improve it’s understanding.
A similar but different concept in terms of the data is being used in the medical space, where AI can read and interpret initial scans and triage. AI has been proven to be more accurate at reading scans consistently, flagging potentially issues to skilled consultants. This can again reduce case load when you are confident of the AI’s capability and where the patient has opted in.
These are examples where AI can help in a sector where skills are in short supply but it’s also creating roles in the training space.
The same is true around bringing more diversity within the models. AI has been around a long time, at KTSL we’ve been using it for 14 years and most of the early AI models were built and design by IT professionals. These meant the mathematical models were fed and designed by one subset of the population. The only actor it knew was that of someone in IT. As it broadens it capabilities it needs to broaden it’s understanding of the ‘actor’s’ it needs to be and that requires training.
Kathryn Milliner
Why has equity and bias become more important in AI?
Aaron Perrott
I think this comes back to the point around data. You will always get a level of bias because it can only understand based on the data it has and the training it’s been given.
As a human, we will listen to somebody’s opinion, but we will use our own judgement about what we take on board and/or what we don’t trust or believe. In essence we can disagree, or take on parts of an opinion based on our own personal perspectives and our own experiences. You could argue that this is also a form of bias and it’s hard to change this, AI has a similar bias.
The difference though, is our bias is built up over our own learnt experience, whereas AI is based upon the data of which it’s been fed. It therefore becomes much easier to force bias into AI. If you want something to be believed. If you tell it enough, it will push that out and it’s a real issue, particularly in the public AI space. You could easily flood commercial public AI with misinformation, because it doesn’t have an opinion and it will push that information out to everybody that consumes it.
The more we rely on it, and the more individuals want to deliberately force bias into AI the harder it is to control the flow of misinformation.
This is why if you’re using a public AI, you should always do a level of follow up and fact checking anything which you need to rely on. An AI based on workplace data is different because it’s only taking data from your environment which would have a level of checking already.
Kathryn Milliner
On that topic, a colleague and I both asked co-pilot the same question slightly different way and it gave us the same answer but in reverse to support our argument.
That question being is Die Hard a Christmas movie? I said, It’s not and co-pilot argued with a set of examples about why it’s not a Christmas movie. My colleague then fed the question but in terms of saying why it is and it used exactly the same reasons to explain why it was a Christmas movie.
It’s an example with a silly subject and how it’s subjective in it’s response and biased based on how it was asked.
Aaron Perrott
Very good example. This is all down to one of the earlier concepts of a user prompt. Over time AI has a very human quirk to it, in that often wants to provide the answer the person’s expecting. Ie: it will default to the prompt.
Kathryn Milliner
People pleasing!
Aaron Perrott
It has that human like trait, and because it doesn’t really have it’s own opinion it doesn’t argue! Ie: it’s not an opinionated answer. It will only give you an answer which you’ve prompted it for, therefore a level of human induced bias.
Kathryn Milliner
How do you think AI is going to impact women and diversity in talent?
Aaron Perrott
I think there’s a lot of positives here. There is a big opportunity for it to open up workplaces that have been historically more closed to certain gender roles because it breaks down some barriers. In particular it can change or adapt to learning style.
We all know that that people learn differently. Some people like very factual based learning, some prefer more practical based learning, some like more face to face but each style is different.
The cost of producing learning material to cope with this has historically been time consuming and difficult but AI can breakdown these initial barriers.
If we look at IT which is the industry I know best. There’s definitely a shift where being able to explain something in a non technical way or to different types of personality can really help to convey the right level of understanding.
AI allows people coming into the industry to have so much access to information, taught and explained in a way specific to them and how they want to consume it. As opposed to a generic approach defined by one personality type. We are seeing a steady increase in woman moving into STEM based subjects and careers.
It’s also forcing organisations to change. You can come into an organisation now without knowing a huge amount about it, but understand really quickly its values, the way it behaves because AI can expose that quickly. There is no excuse not to adopt differing communication styles and businesses won’t attract the top talent unless they get this right from the start of the recruitment and onboarding processes.
Kathryn Milliner
What would your advice be for leaders, teams and individuals who are leading businesses, where they’re looking at AI or looking at how they can embrace AI within an organisation and in their wider life?
Aaron Perrott
First of all, you do need to embrace it. Regardless of your opinion on it, it is here to stay. Just like every industrial revolution we’ve had in the past you need to work with it.
I would make sure everyone is aware that they should be using a workplace AI and not a public AI for work and they understand the differences and the reasons. As discussed you can’t rely on public AI, particularly in the workplace, and you certainly shouldn’t be sending workplace data into a public AI. This is because once you enter it into a public model, that data is re-used and potentially becomes publicly available if that question is asked about your organisation.
It’s also important to make sure you’re using the right AI for the right job. Each individual organisation is going to have its own data and it’s own focus and you need to ensure it’s accurate for your business.
You also need to put governance around it and make sure you have a governance team for AI internally that’s taking into consideration all of the impact. This is both the technical and the business impact.
For example any cultural barriers to allowing AI to start to use the data and what are the business challenges you are trying to solve.
Back to the one of the earlier points, as you use AI to remove the repetition you need to ensure you are still bringing/training and teaching staff the right experience they need to later on interpret it. If you don’t address that, we could be in a position in 3-5 years time where you don’t have people that can now do the more difficult aspects of the job because no one has come through that system.
It’s important to start to think about how is this going to affect my business and how am I going to embrace it? Because otherwise your business may fall behind the curve. I think the pace of AI will force this, particularly in IT.
My final advice would be to go and learn about it. There is a wealth of good courses and material about how AI works. From good prompt engineering courses to the underlying concepts. Understanding it, is a great way to work out how to best leverage it.