Are AI bots really as smart as we are? A look behind the technology

For a long time, chatbots answered users' questions much like the service staff of companies or public institutions. But rule-based chatbots are now out – artificial intelligence (AI) is on everyone's lips.

But are AI language models already as intelligent as humans? Bots have long since overtaken us at many tasks, yet there's still a lot they don't understand. Or do they simply tick differently from us?

In this post, we take a look at:

• how chatbots have changed,
• how AI bots communicate,
• how alignment research clears problems out of the way,
• how humans and AI become a dream team

‍

Rule-based is a thing of the past

Rule-based chatbots? Old hat. We've come much further: thanks to machine learning, AI-based chatbots get even smarter with every conversation and can answer complex questions too.

Chatbots offer many ways to integrate with other systems via interfaces (APIs). Through these interfaces, they can access additional information and support users even better. One example is connecting to logistics software to provide information on the status of a delivery. The chatbot retrieves the relevant data via the API (application programming interface) and delivers the requested information to users quickly and efficiently. This integration opens up many possible uses and lets the chatbot be embedded seamlessly into existing systems.

‍

How does this communication work?

AI chatbot communication is based on a set of algorithms and models that let them generate human-like answers. A common method is machine learning, in which the chatbot is trained on large amounts of text data to recognize patterns and relationships. That lets it respond to user input and provide meaningful answers.

One example of such a chatbot is GPT-3, which can answer complex questions and even write creative texts. Besides text understanding, chatbots also use other techniques such as natural language processing (NLP) and sentiment analysis to better understand the context of a conversation and respond appropriately. However, AI chatbots aren't perfect yet and can sometimes give wrong or inappropriate answers, so more research and development is needed to further improve their capabilities.

‍

ChatGPT made big waves

The debate about machines was just calming down when ChatGPT came along and reset everything. Reports on job losses and on whether AI will now become smarter than us followed. Conversations with chatbots even made people believe they were dealing with a living system.

But how can chatbots be so good even though they don't understand a word? "They're so good because they're spared the understanding," says Elena Esposito, professor of the sociology of communication at the universities of Bologna and Bielefeld. "Understanding means being able to form abstractions and understand sentences in context – and that takes a lot of effort."

‍

How does language work without understanding?

In the past, a lot of time and effort went into teaching language processing systems grammar and rules. But only when people started training these systems on large amounts of data was there a clear leap in quality. A well-known quote from IBM engineer Frederick Jelinek sums up this process nicely: "Every time I fire a linguist, the performance of the speech recognizer goes up." This illustrates the paradigm shift toward data-driven approaches in language processing.

Today, computer systems trained on extensive datasets can achieve an impressive ability to recognize and process language. This development lets copywriters use their creativity on a new level and make content even more precise and effective.

‍

Is "artificial intelligence" an accurate term?

Experts have closely observed where current chatbots often fail, and their findings support Elena Esposito's thesis. It's not just that they sometimes deliver wrong results – they also regularly fail at logic puzzles, abstract and conceptual thinking and understanding causal relationships.

Some argue that this can be solved and is just teething trouble, while others say these are fundamental weaknesses of machine learning methods that can't be fixed. So there's still a lot to do to develop chatbots further and overcome these challenges.

The researcher would prefer it if we stopped talking about artificial intelligence altogether. "It's not about imitating humans – that idea only causes confusion and fear." Instead, we should focus on the role of machines in communication processes: "These systems are made for communication – and you can do lots of great things with them!"

Some people misunderstand that, though – in one case, someone chatted with Microsoft's Bing chatbot about life's big questions and used it as a relationship adviser. Microsoft then limited the interaction time with the system – but only its communication, of course, not its "intelligence", says Esposito.

‍

The alignment problem

Alignment is an important aspect of communication between people. It refers to how conversation partners attune to each other and find a shared rhythm in how they move and speak. But with the language models behind chatbots, things are different.

They achieve their results in a different way from humans. That's where the alignment problem comes from: the models have to be trained to understand context and generate appropriate answers. Various problems can arise, such as misunderstandings or inappropriate responses to certain requests. Good alignment between the model and the human user is therefore essential for successful communication with chatbots. Achieving this requires continuous improvement and adaptation to users' needs.

‍

OpenAI gives insight into its alignment research

Philosopher Nick Bostrom has talked about the dangers of superintelligence and how AI safety research should deal with them – risks that can come from AI itself or from human misuse.

US AI company OpenAI has taken up Bostrom's demands with its alignment research. Its principle is clear: humans and AI should be brought into harmony – AI learns from feedback and helps people in many areas.

OpenAI wants to make full use of current alignment approaches to explore and document all possibilities and limits. That's why its research on training AI systems rests on these three pillars:

• human feedback
• supporting human evaluation
• conducting alignment research

‍

Rethinking was already needed in Johannes Gutenberg's day

Today's "problems" are by no means new. Writing revolutionized how people remember and think. The arrival of printing and mass media set further changes in motion.

But now we're on the threshold of a new era: learning systems. These systems can extract far more information from data than was ever possible before. That raises the exciting question of how our thinking will change as we increasingly rely on their results.

Will our cognitive abilities be expanded or even transformed? It remains to be seen what impact this development will have on our society and what new opportunities it will open up, for example for copywriters and authors.

‍

Humans and AI as a dream team

To bring humans and machines together in the best possible way today, experts such as melibo are needed. We bring live chat agents and AI chatbots together on one platform. That way, you guarantee your customers the best service.

AI helps in live chat by suggesting suitable answers to agents. At the same time, the chatbot can be trained so that over time it answers more and more questions itself and live chat agents have to answer fewer.

‍

Conclusion: communication between humans and machines can work brilliantly and bring many benefits – if we learn to understand it properly

There have always been challenges – even Johannes Gutenberg had to struggle before his printing press earned him nicknames like "trigger of a media revolution" and "man of the millennium".

So there was no standing still back then either. The rapid pace of digitalization has led to continuous progress in many areas, including artificial intelligence (AI). The interplay of humans and AI brings many exciting opportunities and challenges. For example, AI can be used in customer service to offer automated answers to frequently asked questions or carry out complex data analyses. In medicine, too, AI can help with diagnoses or with discovering new treatments.

However, it's important that these technological advances are used responsibly and that people still play the essential role. melibo is ready as an expert in this field to enable humans and AI to work together harmoniously and so gain competitive advantages. Because the competition never sleeps!

We help you better understand how AI chatbots tick, what value they bring you and your customers and how you can leave your competition behind! What are you waiting for?