How Making AI More Human Will Revolutionise Work
Cognitive analytics, a term that to the uninitiated might sound like some new behavioural therapy technique, is one of the core concepts that underpins machine learning and deep learning systems crucial to the analytical applications across industries.
Yet as is the goal with AI more generally, cognitive analytics uses AI and machine learning to process large amounts of data and apply human-like intelligence to tasks.
“The rise of AI now allows tasks to be completed with almost human-like intelligence through cognitive analytics, which is quickly becoming the expectation for more advanced systems,” explains Dr. Omri Allouche, Chief Scientist at Gong.
Despite the growth we have seen in the amazing capabilities with the likes of ChatGPT and its abilities to convincingly hold a conversation, this coveted human-like intelligence remains elusive.
Sure, AI systems can look at a picture of a house and recognise that a cat is present within it, yet it may struggle to present a deeper contextual understanding of its presence in the context of the image.
“In order for AI to become “cognitively sound”, we need systems that don’t just understand data—they understand the context, subtleties, and implications behind it,” Adnan Masood, Chief AI Architect at UST.
This step is what we think of when discussing cognitive AI. Luckily, advancements in AI more generally are ushering in abilities that are making AI more and more human-like, bringing in countless new applications.
Humanising AI
Current language models demonstrate impressive capabilities in natural language processing and generation. However, promising developments are being developed that focus on enhancing reasoning abilities.
This includes step-by-step reasoning, causal inference, and logical deduction and induction. These advancements aim to improve models' capacity to break down complex problems, understand cause-and-effect relationships, and draw logical conclusions.
Better training data is one of the causes, which can now be synthetically generated, or is more widely collected and more specific to train the model in a more targeted way. This means better, more specific AI systems.
“If you handed a sales team an AI tool that was based on broad, general LLMs,
they’d be able to generate answers to nearly any question but few would be of any use,” explains Dr Omri. “They would be far better served by one trained on a smaller number of hyper-relevant parameters like real sales calls, as these would be capable of understanding the context of questions being asked to generate meaningful results.”
Equally, algorithmic advances have pushed the boundaries of what's possible. The exponential rise in computing power has already enabled models to be trained on internet-scale data, which has seen AI systems develop advanced cognitive abilities like reasoning, planning and even understanding humour. As computing capacity grows, so does this stand to.
Such computational power will place systems to better perform ‘few-shot’ learning. This gives AI systems the greater ability to learn new things without having been told numerous times before.
“Think about the way a child is taught maths. Instead of expecting them to write down a final answer and nothing else, we teach them to break down their reasoning and show how they came to their calculation,” says Dr Omri. “Modern AI systems are being trained in the same manner, leading to a leap forward in their cognitive abilities.”
This all comes together to help usher in a future of one of the really exciting and truly revolutionary stages of AI: Agentic AI.
Application of Agentic AI
Characterised by its ability to autonomously pursue complex goals and workflows with minimal human supervision, agentic systems represent AI working autonomously.
Unlike traditional AI systems that are programmed for specific tasks, Agentic AI exhibits sophisticated capabilities such as autonomous decision-making, contextual reasoning, adaptable planning, and advanced language understanding.
Agentic AI operates more like a human employee, capable of understanding context, making judgement calls, and efficiently navigating multi-step processes across various applications.
This enhanced level of autonomy and intelligence enables Agentic AI to tackle intricate challenges across diverse domains, potentially transforming how enterprises approach automation and decision-making.
Warnings of Agentic AI
“Reasoning and agentic behaviour will be the next areas of innovation to bear enterprise-deployable solutions,” explains Jeff Lunsford, CEO at Tealium. “We will soon see models that can perform more complex ‘agentic reasoning’ and achieve goals assigned to them.”
In an enterprise setting, having Agentic AI could act as a huge benefit. Currently, AI works to optimise the tasks of humans’ workflow, but Agentic AI would see an independent AI worker on the team.
Imagine a large e-commerce company implementing agentic AI to revolutionise its supply chain management. The AI system, acting as an autonomous agent, can continuously monitor inventory levels, analyse sales trends, and track global shipping conditions.
When it detects a potential stock shortage for a popular product, it doesn't just alert human managers – it takes action. The AI autonomously adjusts order quantities, reroutes shipments to optimise distribution, and could even negotiate with suppliers for better prices and faster delivery times based on external market prices.
Building windows into AI’s machinations
Although the benefits of this are clear to see, what isn’t so clear is understanding the process the AI will take to come to these decisions. Already, arguments are swirling regarding ‘Black Box AI’ regarding even our current AI systems.
“As AI systems advance to the point of making their own decisions, the lines between emergent behaviours and actions we intentionally trained them to do become indistinguishable, making it even more difficult to ensure transparency and accountability,” explains Jeff.
In the context of the previously mentioned ecommerce setting, Agentic AI changed the route of goods so it was longer, yet you can see no particular reason for doing so. Not knowing the internal machinations the system is using to make those decisions means companies may be unsure whether to change course or go with the AI’s choice in the belief it has picked up something and it is not a misinterpretation.
This means Agentic AI becomes increasingly difficult to implement when leveraging it for more regulated industries, like healthcare or finance, where decisions on things like the approval of loans must follow strict guidelines.
“As AI systems become more integral to decision-making, the issues of data privacy and algorithmic bias become even more pressing,” explains Adnan. “These systems must be designed with robust safeguards to ensure they do not inadvertently perpetuate existing biases or violate privacy standards.”
A future of Agentic AI
As we look to the future, the explosion of interest in AI is ushering in advancements that may bring about the age of Agentic AI sooner than many think.
“Imagine a future where a business’ AI agent could contact suppliers. And the ‘person’ who picks up the phone on the other end could also be another AI agent,” explains Dr Omri. “Chatbots talking to each other in some form is inevitable.”
Yet, in the rush of excitement for new AI advancement, Dr Omri argues it is precisely because of this that AI adoption should be accompanied by governance standards and training that addresses what they should and shouldn’t be used for. Already the EU has enact an AI Act, but with new use of AI will come an increasing level of legislation to match.
Looking to the future, the evolution of cognitive analytics and Agentic AI promises to revolutionise industries and redefine human-machine interaction. However, this advancement comes with significant challenges that must be addressed.
By balancing ability with responsibility, enterprises, organisations and indeed the wider world stand to benefit from this evolution of AI and a new way of working it can introduce.
To read the full story in the magazine, click HERE.
**************
Make sure you check out the latest edition of AI Magazine and also sign up to our global conference series - Tech & AI LIVE 2024
**************
AI Magazine is a BizClik brand
