What would African artificial intelligence (AI) look like?(非洲的人工智能会是什么样子？) Olubayo Adekanmbi, a Nigerian entrepreneur, reels off(一口气说出) ideas: as fluent in African languages as in English; as comfortable with speech as with text; and lean(精简的，高效的), so it can flourish where chips are scarce and internet patchy(不稳定的，时好时坏的). By contrast(相比之下), frontier labs in Silicon Valley train gargantuan(巨大的，庞大的) models on English-heavy datasets. “The current gameplay for AI globally is completely un-African,” he says.
非洲的人工智能（AI）会是什么样子？尼日利亚企业家奥卢巴约·阿德坎比随口列举了几个想法：既能流利使用非洲语言，也能流利使用英语；既能轻松处理语音，也能轻松处理文本；还要精简，以便在芯片稀缺、网络不稳定的地方也能蓬勃发展。相比之下，硅谷的前沿实验室用大量英语数据集训练庞大的模型。他说：“目前全球AI的玩法完全是非非洲的。”
Which is why(这就是为什么) two years ago, after a career in telecoms, Mr Adekanmbi co-founded EqualyzAI. The startup has designed tools to advise mothers on health, help traders with accounts and handle customer calls for banks, in English and four Nigerian languages. Similar outfits(机构，公司) are springing up(涌现，迅速出现) all over the continent.
这就是为什么两年前，在电信行业工作多年后，阿德坎比先生共同创立了EqualyzAI。这家初创公司设计了多种工具，用英语和四种尼日利亚语言为母亲提供健康建议、帮助商人处理账目、为银行处理客户来电。类似的机构正在整个非洲大陆如雨后春笋般涌现。
They might seem redundant(多余的，冗余的), given the spread(鉴于……的普及) of frontier models. Microsoft estimates that among people of working age 8% of Ethiopians, 9% of Congolese and 23% of South Africans have used generative AI. OpenAI says the number of ChatGPT users in Africa is growing faster than in any other region, albeit(尽管，虽然) from a low base. Chinese models are also popular.
鉴于前沿模型的普及，它们可能看起来多余。微软估计，在适龄劳动人口中，8%的埃塞俄比亚人、9%的刚果人和23%的南非人使用过生成式AI。OpenAI表示，非洲的ChatGPT用户数量增长速度超过其他任何地区，尽管基数较低。中国模型也很受欢迎。