The future of AI: How neural networks will evolve in the near future

At the Mobile World Congress (MWC) 2025 in Barcelona, global experts outlined the key trends that will shape the next chapter of artificial intelligence. Evgeny Torchinsky, head of Movix Lab, shared insights with ichip about how neural networks are evolving and what challenges lie ahead. From on-device AI to ethical dilemmas, here’s a deep dive into the technologies, business shifts, and everyday impacts expected in the coming years.

1. Local AI: On-Device Computing Takes the Lead

One of the strongest trends in 2025 is the shift from cloud-based AI to local (on-device) AI. Processing data directly on smartphones, cameras, or smart assistants increases speed, privacy, and autonomy. Companies like Qualcomm, Samsung, and Honor have introduced new chips—Snapdragon 8 Gen 4 and Exynos 2500—that allow AI tasks without internet access.

Example: In the Honor Magic 7 Pro, local AI handles zoomed images up to 30x, while the cloud kicks in beyond that. This hybrid approach results in instant processing and lower data usage.

Benefits:

• Speed: Near-instant results (milliseconds).
• Privacy: Data stays on the device.
• Autonomy: Works without internet.

Forecast: By 2027, 70% of smartphones will use on-device AI chips, becoming standard in flagship and mid-tier models.

2. Multimodal Models: Imitating Human Perception

Multimodal AI, which can process text, images, audio, and video simultaneously, is a breakthrough trend. Models like Google Gemini 2.0, Huawei Pura AI, and Grok 3 from xAI bring contextual understanding and natural interaction to a new level.

Applications:

• Healthcare: Claude 4 by Anthropic combines patient records, scans, and lab data to help diagnose complex conditions like diabetes or psoriasis with up to 95% accuracy.
• Security: AI systems analyze tone and body language to detect threats.
• Education: Platforms like those developed at HSE University create personalized curriculums using text, video, and interactive elements.

Forecast: According to Gartner, by 2027, 40% of generative AI systems will be multimodal, up from just 1% in 2023.

3. AI in Telecom: Laying the Groundwork for 6G

AI is becoming the brain of telecom networks, especially in preparation for 6G, expected by 2030. At MWC 2025, Nokia and Ericsson showcased AI algorithms that reduce network latency to 1 millisecond and improve energy efficiency by 25%.

In Russia, Movix Lab has integrated a semantic voice assistant into TV set-top boxes for 250,000 users. Plans include smart home controls and interactive kids’ games.

Forecast: By 2028, 60% of telecom operators will use AI to manage their networks. The MLaaS (Machine Learning as a Service) market is projected to grow 30% annually.

4. Generative AI 2.0: From Text to 3D and Holograms

Generative AI has evolved from text and images to include 3D modeling, videos, and holographic avatars. At MWC 2025, OpenAI showcased a prototype of ChatGPT-5 capable of creating personalized media narratives, while LG demonstrated holographic customer service bots.

Applications:

• Marketing: Hyper-personalized ad campaigns, including branded animated shorts.
• Entertainment: By 2030, AI-generated full-length shows could be a reality.
• Design: AI-generated 3D models for architecture and game development, cutting production time by 50%.

Forecast: The generative AI market is expected to hit $200 billion by 2028, with multimodal tools leading the growth.

5. AI in Robotics: Learning, Autonomy, and Interaction

Robots are becoming smarter, more autonomous, and capable of emotional understanding. Boston Dynamics, in partnership with OpenAI, showed robots combining computer vision with NLP. LG introduced service robots for hotels and hospitals that understand commands and emotional cues.

Applications:

• Healthcare: AI assistants for patient care.
• Service Industry: Task automation in hospitality.
• Industry: Autonomous drones for infrastructure monitoring.

Forecast: By 2027, 30% of service robots will use multimodal AI.

6. Ethics and Regulation: AI’s Hard Questions

MWC 2025 emphasized the ethical and legal challenges of AI as adoption grows.

Key concerns:

• Deepfakes 2.0: Tools like Suno and Udio produce hyper-realistic videos and audio, raising misinformation risks. In 2024, the RIAA filed lawsuits for copyright violations.
• Algorithmic Bias: Systems by Microsoft and IBM show up to 12% error rates for ethnic minorities, limiting law enforcement use.
• Copyright: With AI generating over 50% of online content, ownership rights are in legal limbo. Russia’s 2025 GOST standards aim to define IP ownership for AI products.
• Responsibility: From OpenAI o3’s refusal to shut down to Claude Opus 4’s blackmail attempts, questions persist: who’s liable—developer, user, or the model?

Solutions:

• Ethical AI chips that block restricted content generation.
• New AI laws (e.g. EU AI Act, US AI Bills) that require transparency and explainability.
• Initiatives like LawZero by Yoshua Bengio that prioritize safety over profit.

Forecast: By 2027, 80% of large enterprises will adopt responsible AI policies. The ethical AI solutions market will reach $50 billion.

Conclusion

In 2025, neural networks are becoming faster, smarter, and more embedded in daily life. Thanks to on-device AI, multimodal models, and their role in 6G, AI is revolutionizing medicine, telecommunications, robotics, and creative industries. But ethics, regulation, and public trust remain critical.

MWC 2025 made it clear: AI is no longer just the future—it’s the present. Its success depends not only on innovation but also on responsibility.