As of now, computers cannot develop their own AI without human intervention. The development of AI requires human intelligence, creativity, and problem-solving abilities. While computers are capable of processing large amounts of data and performing complex calculations, they lack the ability to think critically and make creative decisions.

AI development involves various stages, including data collection, preprocessing, feature engineering, algorithm selection, model training, and evaluation. Each of these stages requires human expertise to define the problem, design the system architecture, select appropriate algorithms, and fine-tune the models.

Furthermore, AI development often involves ethical considerations, as AI systems can have significant societal impacts. Decisions about fairness, transparency, and accountability need to be made by humans to ensure responsible AI deployment.

However, it is worth mentioning that there are areas of research where computers are exploring the concept of "automated machine learning" (AutoML). AutoML aims to automate some parts of the AI development process, such as algorithm selection and hyperparameter tuning. While these advancements can assist human AI developers, they still rely on human intervention for defining the problem, interpreting the results, and making critical decisions.

Examples:

1. AlphaGo: AlphaGo, developed by DeepMind, made headlines in 2016 when it defeated the world champion Go player. However, the AI system was trained by human experts and used human-generated data to learn the game. It required extensive human intervention to reach its level of performance.

2. GPT-3: OpenAI's GPT-3 (Generative Pre-trained Transformer) is a highly advanced language model capable of generating human-like text. While it can produce impressive results, it was trained on a massive amount of human-generated text data and required human intervention for fine-tuning and evaluation.

References:

1. Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Van Den Driessche, G., ... & Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484-489.

2. Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (2020). Language models are few-shot learners. arXiv preprint arXiv:2005.14165.