‘Lovelock places great emphasis on proof…He is concerned that projections are relying on computer models…because models of that kind have let us down before’