CEO of D-Gn Johanna Cabildo about the reason for the start of the most intelligent artificial intelligence with better data

In a recent conversation with Johanna Cabildo, co -founder and executive director D-gnWe dug why the future of artificial intelligence may depend on algorithms and more on those behind the data.

While a lot of industry focuses on formatting form or agents based on Blockchain, D-Gn is betting on something more fundamental: building high-quality and marks of society to make artificial intelligence more clear, faster and better.

In our chat, Cabildo shared how the best data leads to better models, why decentralization on property rights does not matter, and how D-Gn helps companies gain a real advantage in artificial intelligence race.

Your team emphasizes that the most intelligent artificial intelligence begins with better data, not larger models. How do D-Gn focus on high-quality training data as its basic mission?

As you know, everyone chases the next hacking model, but they miss the forest of the trees. We have realized early that the most advanced algorithm in the world is only good as you feed. Think about it, do you prefer to have a wonderful chef that works with spoiled ingredients or good chefs with the finest fresh ingredients?

The real transformation point came during my work in the Aramco X DropPgroup AI project. We have seen DropGroup models outperform giants like Openai, not because we had more account, but because our data pipelines were simply better: more accurate, more organized and designed for this purpose. It became painful painfully that even the best players were under pressure to cut the corners, ignore the Internet, chase the range and sacrifice the speed. Their models were fast, but their results were often biased, inconsistent and expensive for operation.

That is when it strikes me, we do not need the larger graphics processing units, we need better data. Better set. Better structure. Better delivery to the developer of artificial intelligence. This is the origin of D-Gn. It is not just the other Amnesty International, but the data infrastructure that makes all artificial intelligence more intelligent, faster and more worthy of confidence.

Many artificial intelligence companies are accurately, but a lack of role in the role. Can you explain what the “role” training means in practice and why does it matter in the AI?

It is similar to training in the role of the difference between the general practitioner and the heart surgeon. Both are doctors, but you want a specialist when you depend on your life.

Take our work with independent vehicles – the general organism recognition may determine 85 % of the time. But the guardians of our data are not only described by “Stop Sign”. They suspend regional differences, weather conditions, sabotage patterns, and vision corners. When this data is trained AI model, it only sees stop signs, it understands it in the context. This is the difference between 85 % accuracy and 99 % accuracy, which in independent driving, literally saves life.

For institutions, this means that artificial intelligence does not only work in perfect conditions, but also works in a real, chaotic, complex world where we already live.

In a place obsessed with independent factors and coordination, D-Gn sorted a unique position. Why did you deliberately go to the class noise?

Because agents are only smart like their basis. It is like skyscrapers in every person on fast sand and wondering why they continue to fall.

The agent layer is exciting – get it. But what is the value of the decisions to make an independent agent on the basis of biased, incomplete or explicit data? The full promise of the economy is the agents that are not associated with smart and smart efficiency. You cannot achieve this if the basic data is contaminated by bad methods or ethics at risk. We decided to solve the founding problem first. Once you get artificial intelligence that really understands the world, you can unleash to act independently.

We are not fighting. We are supporters of writing. And real intelligence begins with a proper understanding of reality, which begins with data that accurately reflects reality.

D-GN data sets are designed to make artificial intelligence more intelligent and more efficient. Can you share examples of customers who use this data to improve the task results?

I can’t name certain customers because of NDAS, but I can share the impact patterns we see. A single digital human platform witnessed a simultaneous accuracy of the lips from 72 % to 99.2 % after training on the continuous facial movement data set. This is the difference between the trick of the sunken valley and a virtual assistant that can be believed to depend on a real person.

We had an emotional company of artificial intelligence that reduces its cross -cultural interpretation errors by 73 % using emotional expression data in the world. In customer service applications, not only the wrong reading emotions create embarrassing reactions – they destroy confidence and can unnecessarily climb conflicts.

The Game Studio, which works on the next generation of NPCS, increased 95 % of behavioral reality after the implementation of our continuous learning pipeline for human accurate expressions and contextual responses. Without this type of living data, their characters felt the mechanism and the players were unable to keep indulging.

The pattern is fixed: companies come to us when they are stuck in a way of anxiety, such as saying with digital humans working in experimental offers but they fail in the real world scenarios. Emotional and behavioral data groups that are constantly identified to learn continuously help them to move to actual commercial publishing-the difference between the grandmother and the tool that people actually need to use.

Many of the public’s attention still focuses on typical architecture. What do you think is tolerated when it comes to the actual material of Amnesty International: data?

The entire conversation on the ethics of artificial intelligence, bias and safety is a conversation about data quality, but no one wants to admit it.

You can build a more elegant and sophisticated model structure in the world, but if you train it on the biased data, you will get a biased AI. If you train it on incomplete data, you will get the fragility of artificial intelligence. If you train it on or recycled artificial data, you will get increasingly separate from reality.

What is ignored is that data is not just a fuel for Amnesty International – it’s DNA. It not only determines what artificial intelligence can do, but who serves, how he behaves and whether it makes the world actually better or only inflated the problems exist on a large scale.

In 2025, many organizations are trying to control artificial intelligence for very specific tasks. How are your team building data sets that are not only clean, but are rich in context and adaptable to fine roles?

We have built what we call the “human being in the episode”, but not the way most people think about it. Parents are not only determined on data-they are cultural translators, context of context and edge.

When we build a data collection, for example, moderate content, we only know “inappropriate content”. Our shareholders define the cultural nuances, the suitability based on context, the advanced colloquial appendix, and regional sensitivities. Amnesty International not only learns the rules – it learns the exact art of human rule.

Quality Assurance Points System (QAS) guarantees that shareholders who have experience in the deep field in specific areas are data groups for these fields. The legal expert, a legal expert, is legal content, the original speakers deal with multi -language challenges.

What is the infrastructure or tools needed to support high-resolution D-GN goals, and how do you maintain this quality on a large scale?

We have built our entire infrastructure on Blockchain for some reason – unchangeable audit paths. Each explanation, each quality degree, each procedure is registered permanently. You can track any part of the data named to those who created it, and when and with any quality standards.

Our dynamic discovery system uses artificial intelligence to inform abnormal cases in actual time, but humans make final calls. We have operated the process – shareholders get reputation, open better tasks, and join elite teams. The quality is not only required, but it is rewarded and celebrated.

The main insight is that the scale without quality is just average scale. We prefer that we have 10,000 highly skilled and stimulating shareholders of only 100,000 clicks through tasks such as robots.

I previously talked about making artificial intelligence development more fair. How does your data approach help convert energy away from central players and towards a more diverse environmental system?

Data is the power, and for a very student, that force is concentrated in the hands of a few technology giants who can withstand the integration of the entire Internet. We create democratic data.

Parents keep our data for ownership classes in the data groups they help create. When the institution licenses a set of data, shareholders get continuous royalties in USDT. For the first time, people who train artificial intelligence systems who train their intelligence and effort benefit from the success of artificial intelligence.

More importantly, we are creating data representing various views, cultures, contexts-not only what municipality mockery of web sites in English. This means that artificial intelligence trained on our data actually works for everyone, not just seeing the silicon valley.

What do you think of the current organizational batch about the transparency and safety of artificial intelligence? How can the best training data help companies to survive and moral?

Organizers ask the correct questions, but most artificial intelligence companies cannot answer because they do not know what is in their training data. They ignored it from unknown sources, and dealt with them through black boxes and hoped for the best.

Our ONSAIN approach means that each part of the training data has a source. Companies can prove that artificial intelligence has not been trained in copyright -protected materials, biased data groups or questionable sources. They can show compliance not promises, but with fixed proof.

Better training data is not only related to better performance – but rather to build artificial intelligence that you can already trust and organize. I see a world in which the organizers are asking for all artificial intelligence models from GPT to GROK.

Finally, with a lot of noise in artificial intelligence space, what is the wrong concept about “training data” that you want to wipe it once and for all?

More data is always equal to the best Amnesty International. no.

Quality excels the quantity every time. A thousand examples rich in context, which will excel rich in context over one million of them. We have seen companies spending millions on huge data groups that make artificial intelligence already worse because the data were noisy, biased or wrong. As they say in computing, garbage in, garbage outside.

But this is the real discriminatory factor that everyone lacks: data variations. Think about it logically for one second – if we all train artificial intelligence models on the same homogeneous data collections, what do you get? The same predictive outputs. This does not lead to the human race forward or advances in the race of artificial intelligence.

Data differences-edge cases, cultural differences, and chaos in the real world-and this separates the best models in their category from medium models. It is the difference between artificial intelligence that works in controlled environments and AI that flourish in the chaos of the actual human experience.

The future belongs to Amnesty International trained in smaller, smarter and humanitarian data groups with rich-rich-and-larger differences, which flattens human complexity in the similarity of Khwarizmi. This is not only our business model, it is our mission, which makes artificial intelligence already smart, and not only impressive.

In D-Gn, we build an environmental system for partner organizations that share this vision-stakeholders who live before “Amnesty International in favor of humanityGiven that the human data created, rich in context, which is designed for this purpose with the differences in the real world, is not only the best-it’s the only path for Amnesty International that we can already be trusted in important decisions.

Read more Interviews here.