As AI continues to advance, businesses are exploring new ways to leverage the technology to tackle their biggest challenges. One promising application is hyperautomation, an AI-driven approach to automating manual processes that has recently gained significant traction in the business world.
To help you learn more about hyperautomation and how it will shape the future of work, we talked to Singapore-based Kishan S.—Managing Director at Analytico Asia and Lead Data Science Instructor at General Assembly—and Australia-based Greg Baker—Director at the Institute for Open Systems Technologies.
Or read on for highlights discussing:
- What is hyperautomation and what are the challenges with automating complex workflows
- Why domain expertise is one of the most important hard skills in a hyperautomated workplace
- How close AI technology is to delivering full hyperautomation
What is hyperautomation?
Kishan: Hyperautomation is the process of collecting and cleansing data to build predictive and/or analytical models that automate complex tasks. Basically, it’s a way to completely eliminate a manual task such as data entry, invoice processing, or even customer service responses.
This allows businesses to streamline operations, reduce human error, and increase efficiency at scale. Although full hyperautomation is not yet possible, we are getting closer to making it a reality.
What are the benefits of hyperautomation?
Greg: Most studies agree that automation can boost productivity for white-collar workers by at least 25%. And that’s just with GPT-4. As we get into smarter and more advanced models, you can expect those numbers to increase exponentially.
Still, the big challenge with automating complex workflows is that it requires processing long sequences of text and that’s still very difficult for LLMs. You see, LLMs have limited context, so they’re better at handling shorter sequences. As sequences grow longer, maintaining coherence becomes more difficult and models struggle to accurately remember earlier parts of the sequence.
Compounding the issue is that once an LLM makes a mistake, it tends to continue on that path, making progressively worse mistakes. All this results in outputs that are inconsistent or just completely disconnected from the initial input. So, until we develop ways for LLMs to effectively handle long sequences, hyperautomation capabilities will be limited, and the process will require lots of human oversight.
What skills will you need in a hyperautomated world?
Kishan: Beyond the basic technical competencies required to work with AI and LLMs, such as coding with Python or understanding statistics, one of the most important hard skills in a hyper-automated world will be domain knowledge. Regardless of the sector you’re in, having a solid understanding of the big picture will be crucial for:
- Identifying the right elements to integrate into a given model.
- Understanding how those elements will influence a model’s outputs.
- Analyzing the potential long and short-term consequences of integrating a model’s outputs.
In terms of soft skills, resourcefulness and adaptability are key. Things are moving fast, and the shelf life of hard skills is ever shrinking. People need to be able to think on their feet and quickly adjust to navigating new environments and challenges. In addition, as AI evolves, a basic understanding of ethics and regulations will be crucial to ensure responsible applications and to build awareness of how to safeguard against misuse.
Which industries are leading the way in hyperautomation?
Greg: In Australia, we’re seeing three sectors with widespread adoption:
- Programming: Programmers across sectors are seeing huge productivity gains with automation. For example, I’ve written over 25,000 lines of code just over the last month with the help of three AI models.
- Government: This one was a big surprise since the government isn’t traditionally an early adopter of most technologies. But, with so much bureaucracy and manual work, the efficiency gains and cost-savings are too good to pass up.
- Translation: Unless you’re highly specialized in a particular area, translation is not a viable career path anymore. So much of it has been automated that the role now requires little to no human input.
Based on these examples, you can see how hyperautomation will lead to scenarios where the job is no longer to do the thing but to oversee the bots doing those things.
Kishan: While the market for AI technology in Singapore is still not on par with that of Australia or the European Union, both the public and the private sectors have made significant investments in AI and automation.
Since the government has been so proactive about investing in the technology, we’ve experienced a surge in data and AI governance roles focused on addressing issues such as: Who governs model training? What happens when a model fails? Who owns the data on which models are trained? And what are the risks involved?
How close are we to full hyperautomation?
Greg: I’ll cite answers from two different sources. The first is from Leopold Ashton Brenner, who used to work at OpenAI. He states that full hyperautomation will be possible in around two years.
Now, if I project the number of pre-parameters in the language model over time—which has been on an almost perfect exponential curve—it’s currently on target to exceed the number of synapses in the human brain by 2029. So my answer is somewhere between 2026 at the earliest and 2029 the latest. But, most people in the industry say it’s 2031, while others push it as late as 2035.
Kishan: I think the ability for hyperautomation will vary across sectors and use cases. Some high-performers may achieve hyperautomation by the end of this decade as Greg predicts, but there’s a good chance many will lag behind. There are a lot of factors that affect automation and the quality and capabilities of a model, such as the quality and volume of the data they’re trained on, the algorithms powering them, governance, etc. And in some cases, those factors do not converge, which means there may be a need for manual intervention and fine-tuning.
For example, most AI models are trained on Western data, which creates limitations for use cases not related or applicable to those places and cultures. And then there’s the human element: training professionals, getting them up to a certain level of competence, and then securing the “okays” and the budgets to achieve those goals will take time. So, even if the capabilities are there, we as humans will take a bit longer to catch up.
Which careers are safe from AI replacement?
Greg: One key rule to remember about AI is that it can’t be responsible for something because it cannot own assets. So, in cases where there is a need for ownership and accountability for an outcome, you will always need to have a human involved somewhere in the process. That includes scenarios like legal proceedings, running a company, or being a product manager.
Another thing to keep in mind is that AI will not eliminate all jobs, it will simply enable people to do them better and faster. And I’ll use software development to illustrate the point. Thanks to AI, we can now write programs that would otherwise take years to complete in a very short period of time. But there still needs to be a person responsible for oversight, testing, QA, and delivery. And this is all great news for newcomers and students at places like General Assembly because they can now be incredibly productive with just their bootcamp education and some good prompt engineering.
Kishan: I agree that accountability will be a deciding factor in how much we can automate a job. For now, I think hyperautomation will be limited to non-critical applications—think creating music instead of making healthcare decisions. Still, the technology is moving fast and I believe we should all work on acquiring some AI skills. It’s better to be the person in charge of maintaining the system performing the task than to be the person whose job can easily be replaced by AI.
Make it real
At General Assembly, we deliver the goods to keep you ahead of the curve. Our students are thinkers and doers who don’t wait for some imagined future, but build their skills (AI and more) to contribute to the future they want and need. What we offer is a charge-up from the inside out—so change never stops you in your tracks.
Explore our course catalog and move forward with real skills.