By India Dowley

The swift evolution of artificial intelligence (AI) technologies in recent years has led many to speculate about AI surpassing human capabilities. Some predict AI will take over professional occupations and lead to widespread job loss. However, the reality is more nuanced. While AI has achieved impressive feats in narrow domains, general human-level AI remains firmly outside its grasp. There are profound limits on today's AI technologies that belie dystopian futures envisioned by some.

According to a McKinsey report, just 5% of occupations could be fully automated using current AI capabilities. However, nearly 60% of occupations could have 30% of their activities automated, which will impact jobs. So while portions of jobs may be susceptible, complete human replacement is unrealistic presently.

Here, we take a look at 10 key boundaries facing AI and machine learning today, based on their current capabilities and setups. In short, presently no AI system can effectively replicate the full span of flexible, common sense reasoning and creative abstract thinking exhibited by human minds – but continued research and interdisciplinary approaches will push AI into new frontiers, bringing innovations we cannot yet imagine.

Lack of Common Sense Reasoning

While AI can excel at narrowly defined tasks, it lacks the common sense, intuition and general world knowledge that we acquire over a lifetime. Simple facts like "ice is cold" or complex social dynamics are difficult for AIs to infer without explicit programming. Humans unconsciously apply common sense reasoning to navigate the open-ended challenges of daily life.

Narrow Focus, Not General Intelligence

Today's AI approaches focus on specialised intelligence targeting singular tasks. But human cognition has extraordinarily broad and multifaceted intelligence capabilities. The holy grail of "general AI" that can flexibly reason like a human across different domains remains firmly out of reach presently. AI pioneer Andrew Ng aptly describes today's systems as akin to calculators – extremely capable but only within their specific and limited focus.

Lack of Transparency and Explainability

The inner workings of most AI systems remain largely opaque and difficult to interpret, with decision-making often hidden within black box neural networks. This lack of transparency makes it tremendously difficult to understand why AIs arrived at certain results or ensure algorithmic fairness. Explainable AI is an emerging field trying to make AIs more interpretable and accountable.

Minimal Ability to Transfer Learn

Human learning involves substantial transfer – we can take knowledge learned in one context and seamlessly apply it to entirely new situations. In contrast, knowledge provided to a specific AI model remains isolated within that model. Present AIs must be extensively retrained and redeveloped to transfer learning to new tasks.

Overreliance on Training Data

AI algorithms depend completely on having access to vast amounts of training data. Any biases, distortions or gaps within that data get amplified through the machine learning process. Humans have top-down cognitive abilities with mental models and abstract reasoning to counter biases. Current AIs lack these innate safeguards, emphasising the need for diverse and balanced training datasets.

Language Decontextualisation

Despite advances, most AI still struggles to handle the nuances of broad language context beyond processing simple text snippets. Elements like sarcasm, cultural references, metaphors and subtext are challenging for AIs lacking human experiences and modes of reasoning. Free-form multi-turn dialogue also remains difficult without very focused domains.

Lack of Physical Grounding

Humans learn enormously through interacting with and manipulating the world around us. Yet current AIs are purely software-based and disembodied. Robotics research is trying to bridge this gap but integrating flexible, dexterous robotic skills remains limited. Physical grounding could provide vital learning signals missing from today's AIs.

Deficient Creative Reasoning

The arts have long been seen as a pinnacle of human creativity and imagination. However, today's AIs at best can mimic or remix existing creative work versus generating truly novel innovations. Humans exhibit qualities like improvisation, conceptual leaps and artistic flair that spring from nothing – capacities still absent in AI.

No Social or Emotional Intelligence

Social and emotional intelligence, empathy, compassion and interpersonal skills come instinctively to people but are still foreign concepts to AI systems. Some research tries to recognise emotions through visual and linguistic analysis but human-level emotional cognition is likely decades away if even possible.

Lack of General Purpose

Present AI needs extensive custom data, programming and training tailored for each niche application. But human minds have remarkably adaptable general purpose cognitive capabilities allowing us to intuitively handle open-ended challenges. We can learn any new skill with minimal exposure – a feat still beyond AIs.

Like this? Check out our article about how you can use AI to help grow your business.


Leave a comment