[AI Weekly] Catch voice phishing with your tone and study, even phone calls... Evolution of AI

By analyzing the tone and speaking speed of the other person on the phone, we alert you to the risk of voice phishing, and call people who are putting off studying English first to encourage them to study.

In industrial sites, it acts as the eyes of welding robots and helps them make precise work decisions.

The use of artificial intelligence (AI), which went from answering questions and writing articles, is expanding to areas that deal with human conversation methods, behavior, and physical work.

Last week, a series of technologies applied to everyday life and industry were unveiled, ranging from AI that detects the risk of fraud during phone calls, agents that persuade people, and lightweight models to be mounted on robots.

◇ AI analyzes not only what you say but also your tone of voice

On the 4th, KT[030200] attracted attention by integrating customer protection services and technology into the 'Ansim 365' brand and introducing 'An Ansim 365 Voice Phishing Detection 3.0'.

A notable change is that AI has expanded the scope of examining call content.

In addition to existing context detection, speaker recognition, and deep voice analysis, it also integrates non-verbal expressions such as conversation context, speech speed, and tone of voice to determine whether there is a voice phishing risk.

It is used to analyze not only what the other person is saying, but also the tone of voice and how fast they are speaking.

KT explained that it applied multimodal AI technology that analyzes voice, text, and nonverbal elements together.

If AI determines a currency is risky, it also provides the basis.

We also applied an on-device method that analyzes the original call in real time on the user's terminal rather than transmitting it externally.

We also released the ‘Safe 365 Report’, which allows customers to check their security status and protection details.

It is a service that shows the security status as a score and provides information on the protection received through spam blocking and voice phishing detection.

◇ “Practice just one sentence”… Call first to encourage study

‘Agent Max’, which was unveiled by Weverse Brain, a conversational AI company, on the 3rd, focuses on eliciting human behavior.

It is designed to start a conversation first and induce actual actions, such as learning or purchasing, based on the other person's reaction.

In the demonstration at the time, an AI tutor called CEO Se-won Jo and encouraged him to study English.

When CEO Cho tried to postpone class, saying he was in a meeting, the AI ​​mentioned that he had ended the call the day before, saying he was busy, and encouraged him to take 30 seconds to practice at least one sentence.

When he repeated the English sentences, he felt a sense of accomplishment and promised to call back in the evening.

In an electric bicycle sales demonstration, the installment amount was explained in daily increments to customers who complained about the price burden, and the battery capacities and prices of competing products were also compared.

Discount benefits and ways to adjust the delivery date were also provided to persuade customers to purchase.

The key is 'EDRA', a behavior design engine that sympathizes with the other person's situation, understands the reason for hesitation, and then changes the perspective or suggests small goals.

Weverse Brain plans to expand its business beyond language services to supply AI agents to external companies.

◇ With the ‘eyes’ of a welding robot… AI that makes on-site judgment

NC[036570] AI announced on the 3rd that it has developed 'Backy Vision 3B', a lightweight visual-language model (VLM) that will be responsible for visual recognition of robots and field equipment.

VLM is a model that processes image and language information together.

It can be used to understand the scene captured by the robot's camera and make decisions necessary for work.

Backy Vision 3B is designed to be mounted directly on robots or industrial edge devices with limited computational resources.

According to NC AI, this model recorded higher performance than the six major public models being compared in three global benchmarks that evaluate robot and spatial understanding abilities.

Based on this, NC AI additionally learned a model specialized for welding sites.

The explanation is that the model showed a level of performance that enabled precise field work judgment in the welding benchmark evaluation.

NC AI is developing physical AI for autonomous welding robots with Hanwha Ocean [042660].

We plan to expand the scope of application to the steel, shipbuilding, and defense fields by collaborating with POSCO DX and Hyundai Rotem [064350].