There are assistants and then there are agents. The way they behave is different. Even among agents themselves there are differences. Some agents simply ‘act on someone else’s behalf’ but then there are the kind of agents that ‘take an active role to produce a specified effect’. Think recruitment agent vs talent agent. The latter carries out a function as a satisfactory middle person is expected to. The former goes out of their way to get their client working in the best jobs.
Who would you want helping you – an Assistant or an Agent?
Artificial Intelligence (AI) is usually developed using machine learning or its more powerful subset called Deep Learning. It is thankfully being applied for more and more useful purposes but the way it is delivered to the user is usually through ‘assistance’. Sometimes it’s a personified assistant like Siri, Alexa or Google Assistant. Other times it’s integrated into the platform and as a user you don’t know it’s there but under the hood, your data is being analyzed and classified with AI. For example, Linkedin recently announced it was using AI to match job hunters to jobs on its platform which may not be obvious to the user in the delivery. Another guise you may find AI being used is by Chatbots that operate a ‘conversational AI’ for customer service. In addition, the humble assistant continues to grow its repertoire improving its skills in booking appointments and composing emails.
Are we being too passive with the Assistant approach – do we need something more pushy to do our bidding?
Assistants aim to be personal but this is a hard task given they only take fragments of our preferences over time and try to deduce what we want. There will always be limitations to this approach unless they can obtain the huge amounts of data needed in which they sometimes succeed if the usage is high enough. Being able to tune this accurately to our personal preferences is the real challenge. An assistant by its nature is subservient, waiting for us to signal what we want. It’s unlikely to display the kind of urgency we would desire in times of pressure and competition. An assistant sounds like a luxury, something we could live without. Going without isn’t necessarily a disaster, it just means we’ll probably have a bigger workload. At the end of the day, an assistant as it is most used today still has to be directed whether it’s to set a timer or carry out a command. When we wait for them to be proactive, they more often than not get it wrong. I’ve lost count the number of times an assistant has suggested a non sensical activity leaving me bewildered. Maybe we need a new approach. Something a step up from assistance and more like a superpower – like a talent agent that only has our best interest at heart.
Enter the AI agent
It bids on our behalf, it finds things we can’t find ourselves, it is our advocate. It’s a superpower the way we always expected AI to be. With an agent, I will happily want to tell it what it needs to know because it’s working for me. It’s not blatantly working for the platform. Whilst it isn’t making a cut, somehow it has the motivation to get me the best deal. What things would it do for me? Could it be a true talent agent finding me jobs or opportunities, or a scout sniffing out sales leads, negotiating offers and helping me market myself. And that holy grail – knowing what I need to do before I even think of it. The really trick part is doing those things without a lifetime worth of data about me. For those of us who struggle to get connected with the right people this could be a godsend. Maybe we lack the network or marketing savvy to promote ourselves. For those of us in that boat, an AI agent is a game changer.
What could be the dangers with this? After all if everyone has a powerful AI agent, the advantage would soon be diminished. Like any talent agent, they can do their utmost to advocate for you but at the end of the day, you need to have that talent. The same would hold for an AI agent. Perhaps part of the agent’s job would be to tell you with unashamed clarity, what you need to change to improve your chances. It could be that AI agent, with its machine like, non empathic thinking is the one to tell you that your efforts will be wasted if you continue down this track where a human wouldn’t. This could be a hidden advantage to having a non human agent sidekick. Another danger lurking with an AI agent is the question of bias. If it learns that certain groups end up being more successful, could it steer its client in a way that compromises their beliefs? The AI’s goal is success and sadly, that isn’t always going to be equal or diverse. This is where the design of such an AI agent (to consider diversity) would need to be carefully considered. The ability to mask identity may shield or harm and the agent needs to know when it should apply it. If done correctly, it could help prevent discrimination. If done incorrectly, it would only serve to amplify it.
So how could an AI agent like this be developed?
There are new types of neural networks, new models and methods that could lend itself to an agent approach. But to start with, we need to change the mindset from that of assistance to one of agency. We all know what a big difference a great agent can make compared to an average assistant. The key might be in AI that works by goal setting rather than pattern matching as deep learning is optimized to do today. The technology may take a closer look at the way the brain functions in order to find a new way to achieve a higher level of intelligence. Whatever method by which the development of an AI agent becomes a reality, it’s likely to be the next big objective for Artificial Intelligence to tackle.